[socket] Fast path for TCP_NODELAY bypass lwip_setsockopt overhead

On ESP32 with CONFIG_LWIP_TCPIP_CORE_LOCKING, bypass lwip_setsockopt()
for TCP_NODELAY by directly modifying tcp_pcb->flags under the TCPIP
core lock. This eliminates ~1091 bytes of overhead per call (socket
lookups, hook, switch cascade, refcounting) for what is just a single
bit flip.

The API frame helper toggles Nagle's algorithm on every message send
via set_nodelay_for_message(), making this a hot path. The fast path
reduces set_nodelay_raw_ from calling the full lwip_setsockopt to just
acquiring the mutex, loading 3 pointers, and flipping the TF_NODELAY
bit.

Only enabled when both USE_LWIP_FAST_SELECT (cached lwip_sock pointer)
and CONFIG_LWIP_TCPIP_CORE_LOCKING (real mutex protection) are
available. Falls back to the standard setsockopt call otherwise.

esphome/components/socket/bsd_sockets_impl.h

@@ -56,6 +56,15 @@ class BSDSocketImpl {
     return ::getsockopt(this->fd_, level, optname, optval, optlen);
   }
   int setsockopt(int level, int optname, const void *optval, socklen_t optlen) {
+#if defined(USE_LWIP_FAST_SELECT) && defined(CONFIG_LWIP_TCPIP_CORE_LOCKING)
+    // Fast path for TCP_NODELAY: directly set the pcb flag under the TCPIP core lock,
+    // bypassing lwip_setsockopt overhead (socket lookups, hook, switch cascade, refcounting).
+    if (level == IPPROTO_TCP && optname == TCP_NODELAY && optlen == sizeof(int)) {
+      LwIPLock lock;
+      if (esphome_lwip_set_nodelay(this->cached_sock_, *reinterpret_cast<const int *>(optval) != 0))
+        return 0;
+    }
+#endif
     return ::setsockopt(this->fd_, level, optname, optval, optlen);
   }
   int listen(int backlog) { return ::listen(this->fd_, backlog); }

esphome/components/socket/lwip_sockets_impl.h

@@ -52,6 +52,15 @@ class LwIPSocketImpl {
     return lwip_getsockopt(this->fd_, level, optname, optval, optlen);
   }
   int setsockopt(int level, int optname, const void *optval, socklen_t optlen) {
+#if defined(USE_LWIP_FAST_SELECT) && defined(CONFIG_LWIP_TCPIP_CORE_LOCKING)
+    // Fast path for TCP_NODELAY: directly set the pcb flag under the TCPIP core lock,
+    // bypassing lwip_setsockopt overhead (socket lookups, hook, switch cascade, refcounting).
+    if (level == IPPROTO_TCP && optname == TCP_NODELAY && optlen == sizeof(int)) {
+      LwIPLock lock;
+      if (esphome_lwip_set_nodelay(this->cached_sock_, *reinterpret_cast<const int *>(optval) != 0))
+        return 0;
+    }
+#endif
     return lwip_setsockopt(this->fd_, level, optname, optval, optlen);
   }
   int listen(int backlog) { return lwip_listen(this->fd_, backlog); }

esphome/core/lwip_fast_select.c

@@ -112,6 +112,7 @@
 // LwIP headers must come first — they define netconn_callback, struct lwip_sock, etc.
 #include <lwip/api.h>
 #include <lwip/priv/sockets_priv.h>
+#include <lwip/tcp.h>
 // FreeRTOS include paths differ: ESP-IDF uses freertos/ prefix, LibreTiny does not
 #ifdef USE_ESP32
 #include <freertos/FreeRTOS.h>
@@ -216,6 +217,17 @@ void esphome_lwip_hook_socket(struct lwip_sock *sock) {
   sock->conn->callback = esphome_socket_event_callback;
 }
 
+bool esphome_lwip_set_nodelay(struct lwip_sock *sock, bool enable) {
+  if (sock == NULL || sock->conn == NULL || sock->conn->pcb.tcp == NULL)
+    return false;
+  if (enable) {
+    tcp_nagle_disable(sock->conn->pcb.tcp);
+  } else {
+    tcp_nagle_enable(sock->conn->pcb.tcp);
+  }
+  return true;
+}
+
 // Wake the main loop from another FreeRTOS task. NOT ISR-safe.
 void esphome_lwip_wake_main_loop(void) {
   TaskHandle_t task = s_main_loop_task;

esphome/core/lwip_fast_select.h

@@ -66,6 +66,13 @@ void esphome_lwip_wake_main_loop(void);
 /// @param px_higher_priority_task_woken Set to pdTRUE if a context switch is needed.
 void esphome_lwip_wake_main_loop_from_isr(int *px_higher_priority_task_woken);
 
+/// Set or clear TCP_NODELAY on a socket's tcp_pcb directly.
+/// Must be called with the TCPIP core lock held (LwIPLock in C++).
+/// This bypasses lwip_setsockopt() overhead (socket lookups, switch cascade,
+/// hooks, refcounting) — just a direct pcb->flags bit set/clear.
+/// Returns true if successful, false if sock/conn/pcb is NULL.
+bool esphome_lwip_set_nodelay(struct lwip_sock *sock, bool enable);
+
 /// Wake the main loop task from any context (ISR, thread, or main loop).
 /// ESP32-only: uses xPortInIsrContext() to detect ISR context.
 /// LibreTiny lacks IRAM_ATTR support needed for ISR-safe paths.