[core] Remove pre-sleep socket scan from fast select path

The pre-sleep scan in Application::yield_with_select_() walks
monitored_sockets_ on every loop iteration, issuing a volatile
cross-thread read on each socket's lwip_sock::rcvevent to preserve
select() semantics when the FreeRTOS task notification counter had
been consumed but a socket still had unread data.

That scenario only existed because of a Socket::ready() contract
violation: callers could stop reading with rcvevent > 0, leaving
data behind with no pending notification. That contract is now
documented and enforced (#15590), and #15589 (the first failure
that reverted the earlier removal attempt #14475) has been fixed.

With the contract honoured, every rcvevent > 0 is paired with a
pending xTaskNotifyGive from the lwip event_callback wrapper (see
lwip_fast_select.c). ulTaskNotifyTake either returns immediately
(counter non-zero) or wakes the moment the notify lands — the scan
has nothing left to rescue.

Evidence: https://github.com/esphome/esphome/pull/15638 — an
instrumentation PR ran across 5 devices (ESP32 rev1/rev3.1/C3 on
Ethernet and WiFi, plus LibreTiny RTL8720CF) through Home Assistant
disconnect/reconnect cycles, multi-client API logger bursts, and
BLE GATT connect storms. Across ~275,000 scans and 4 observed
load-bearing candidates, every hit was in the 2–14µs range — the
instruction-level window between the lwip callback writing
rcvevent and calling xTaskNotifyGive a few instructions later.
Zero hits exceeded 100µs. No hit came anywhere near loop_interval
(16ms), which is the latency scale the scan was added to prevent.

In addition to being unused, the scan is actively harmful on the
hot path: N volatile 16-bit loads against cache-cold cross-thread
lwip_sock structures on every main-loop iteration, just to
reproduce a microsecond-scale ordering artifact the notification
path is already handling authoritatively.

This also removes the now-unused monitored_sockets_ vector and
Application::{register,unregister}_socket() on the fast-select
path. Socket implementations now call esphome_lwip_hook_socket()
directly to install the netconn event callback wrapper.

esphome/components/socket/bsd_sockets_impl.cpp

@@ -14,9 +14,13 @@ BSDSocketImpl::BSDSocketImpl(int fd, bool monitor_loop) {
   if (!monitor_loop || this->fd_ < 0)
     return;
 #ifdef USE_LWIP_FAST_SELECT
-  // Cache lwip_sock pointer and register for monitoring (hooks callback internally)
+  // Cache lwip_sock pointer (used by ready() for direct rcvevent reads) and hook the
+  // netconn event callback so the main loop is notified via FreeRTOS task notifications.
   this->cached_sock_ = esphome_lwip_get_sock(this->fd_);
-  this->loop_monitored_ = App.register_socket(this->cached_sock_);
+  if (this->cached_sock_ != nullptr) {
+    esphome_lwip_hook_socket(this->cached_sock_);
+    this->loop_monitored_ = true;
+  }
 #else
   this->loop_monitored_ = App.register_socket_fd(this->fd_);
 #endif
@@ -30,12 +34,10 @@ BSDSocketImpl::~BSDSocketImpl() {
 
 int BSDSocketImpl::close() {
   if (!this->closed_) {
-    // Unregister before closing to avoid dangling pointer in monitored set
 #ifdef USE_LWIP_FAST_SELECT
-    if (this->loop_monitored_) {
-      App.unregister_socket(this->cached_sock_);
-      this->cached_sock_ = nullptr;
-    }
+    // All LwIP sockets share the same static event_callback, so there is no per-socket
+    // unhook needed — just drop the cached pointer before the socket is destroyed.
+    this->cached_sock_ = nullptr;
 #else
     if (this->loop_monitored_) {
       App.unregister_socket_fd(this->fd_);

esphome/components/socket/lwip_sockets_impl.cpp

@@ -14,9 +14,13 @@ LwIPSocketImpl::LwIPSocketImpl(int fd, bool monitor_loop) {
   if (!monitor_loop || this->fd_ < 0)
     return;
 #ifdef USE_LWIP_FAST_SELECT
-  // Cache lwip_sock pointer and register for monitoring (hooks callback internally)
+  // Cache lwip_sock pointer (used by ready() for direct rcvevent reads) and hook the
+  // netconn event callback so the main loop is notified via FreeRTOS task notifications.
   this->cached_sock_ = esphome_lwip_get_sock(this->fd_);
-  this->loop_monitored_ = App.register_socket(this->cached_sock_);
+  if (this->cached_sock_ != nullptr) {
+    esphome_lwip_hook_socket(this->cached_sock_);
+    this->loop_monitored_ = true;
+  }
 #else
   this->loop_monitored_ = App.register_socket_fd(this->fd_);
 #endif
@@ -30,12 +34,10 @@ LwIPSocketImpl::~LwIPSocketImpl() {
 
 int LwIPSocketImpl::close() {
   if (!this->closed_) {
-    // Unregister before closing to avoid dangling pointer in monitored set
 #ifdef USE_LWIP_FAST_SELECT
-    if (this->loop_monitored_) {
-      App.unregister_socket(this->cached_sock_);
-      this->cached_sock_ = nullptr;
-    }
+    // All LwIP sockets share the same static event_callback, so there is no per-socket
+    // unhook needed — just drop the cached pointer before the socket is destroyed.
+    this->cached_sock_ = nullptr;
 #else
     if (this->loop_monitored_) {
       App.unregister_socket_fd(this->fd_);

esphome/core/application.cpp

@@ -449,32 +449,7 @@ void Application::enable_pending_loops_() {
   }
 }
 
-#ifdef USE_LWIP_FAST_SELECT
-bool Application::register_socket(struct lwip_sock *sock) {
-  // It modifies monitored_sockets_ without locking — must only be called from the main loop.
-  if (sock == nullptr)
-    return false;
-  esphome_lwip_hook_socket(sock);
-  this->monitored_sockets_.push_back(sock);
-  return true;
-}
-
-void Application::unregister_socket(struct lwip_sock *sock) {
-  // It modifies monitored_sockets_ without locking — must only be called from the main loop.
-  for (size_t i = 0; i < this->monitored_sockets_.size(); i++) {
-    if (this->monitored_sockets_[i] != sock)
-      continue;
-
-    // Swap with last element and pop - O(1) removal since order doesn't matter.
-    // No need to unhook the netconn callback — all LwIP sockets share the same
-    // static event_callback, and the socket will be closed by the caller.
-    if (i < this->monitored_sockets_.size() - 1)
-      this->monitored_sockets_[i] = this->monitored_sockets_.back();
-    this->monitored_sockets_.pop_back();
-    return;
-  }
-}
-#elif defined(USE_HOST)
+#ifdef USE_HOST
 bool Application::register_socket_fd(int fd) {
   // WARNING: This function is NOT thread-safe and must only be called from the main loop
   // It modifies socket_fds_ and related variables without locking

esphome/core/application.h

@@ -534,12 +534,7 @@ class Application {
 
   /// Register/unregister a socket to be monitored for read events.
   /// WARNING: These functions are NOT thread-safe. They must only be called from the main loop.
-#ifdef USE_LWIP_FAST_SELECT
-  /// Fast select path: hooks netconn callback and registers for monitoring.
-  /// @return true if registration was successful, false if sock is null
-  bool register_socket(struct lwip_sock *sock);
-  void unregister_socket(struct lwip_sock *sock);
-#elif defined(USE_HOST)
+#ifdef USE_HOST
   /// Fallback select() path: monitors file descriptors.
   /// NOTE: File descriptors >= FD_SETSIZE (typically 10 on ESP) will be rejected with an error.
   /// @return true if registration was successful, false if fd exceeds limits
@@ -653,9 +648,7 @@ class Application {
   //   and active_end_ is incremented
   // - This eliminates branch mispredictions from flag checking in the hot loop
   FixedVector<Component *> looping_components_{};
-#ifdef USE_LWIP_FAST_SELECT
-  std::vector<struct lwip_sock *> monitored_sockets_;  // Cached lwip_sock pointers for direct rcvevent read
-#elif defined(USE_HOST)
+#ifdef USE_HOST
   std::vector<int> socket_fds_;  // Vector of all monitored socket file descriptors
 #endif
 #ifdef USE_HOST
@@ -898,26 +891,16 @@ inline void ESPHOME_ALWAYS_INLINE Application::loop() {
 #ifndef USE_HOST
 inline void ESPHOME_ALWAYS_INLINE Application::yield_with_select_(uint32_t delay_ms) {
 #ifdef USE_LWIP_FAST_SELECT
-  // Fast path (ESP32/LibreTiny): reads rcvevent directly from cached lwip_sock pointers.
-  // Safe because this runs on the main loop which owns socket lifetime (create, read, close).
+  // Fast path (ESP32/LibreTiny): FreeRTOS task notifications posted by the lwip
+  // event_callback wrapper (see lwip_fast_select.c) are the single source of truth for
+  // socket wake-ups. Every NETCONN_EVT_RCVPLUS posts an xTaskNotifyGive, so any notification
+  // that lands between wakes keeps the counter non-zero (next ulTaskNotifyTake returns
+  // immediately) or wakes a blocked Take directly. Also woken by wake_loop_threadsafe()
+  // from background tasks, or timeout.
   if (delay_ms == 0) [[unlikely]] {
     yield();
     return;
   }
-
-  // Check if any socket already has pending data before sleeping.
-  // If a socket still has unread data (rcvevent > 0) but the task notification was already
-  // consumed, ulTaskNotifyTake would block until timeout — adding up to delay_ms latency.
-  // This scan preserves select() semantics: return immediately when any fd is ready.
-  for (struct lwip_sock *sock : this->monitored_sockets_) {
-    if (esphome_lwip_socket_has_data(sock)) {
-      yield();
-      return;
-    }
-  }
-
-  // Sleep with instant wake via FreeRTOS task notification.
-  // Woken by: callback wrapper (socket data), wake_loop_threadsafe() (background tasks), or timeout.
 #endif
   esphome::internal::wakeable_delay(delay_ms);
 }