Merge branch 'inline-micros-esp32' into integration

Re-merge the updated PR (now keeps esp8266 millis/delay out-of-line so
integration's accumulator + custom delay overlay cleanly, while still
inlining yield/micros/millis_64 on esp8266).

esphome/components/esp8266/core.cpp

@@ -15,7 +15,7 @@ extern "C" {
 
 namespace esphome {
 
-// yield() and micros() inlined in hal.h.
+// yield(), micros(), millis_64() inlined in hal.h.
 // Fast accumulator replacement for Arduino's millis() (~3.3 μs via 4× 64-bit
 // multiplies on the LX106). Tracks a running ms counter from 32-bit
 // system_get_time() deltas using pure 32-bit ops. Installed as __wrap_millis
@@ -66,7 +66,6 @@ uint32_t IRAM_ATTR HOT millis() {
   xt_wsr_ps(ps);
   return result;
 }
-uint64_t millis_64() { return Millis64Impl::compute(millis()); }
 // Poll-based delay that avoids ::delay() — Arduino's __delay has an intra-object
 // call to the original millis() that --wrap can't intercept, so calling ::delay()
 // would keep the slow Arduino millis body alive in IRAM. optimistic_yield still

esphome/core/hal.h

@@ -144,13 +144,9 @@ __attribute__((always_inline)) inline uint64_t millis_64() {
 __attribute__((always_inline)) inline void yield() { ::yield(); }
 __attribute__((always_inline)) inline uint32_t micros() { return static_cast<uint32_t>(::micros()); }
 #if defined(USE_ESP8266)
-// delay(), millis(), millis_64() stay out-of-line on this branch (integration):
-// esphome::millis() is the body of __wrap_millis (-Wl,--wrap=millis), so it must
-// remain a real symbol to avoid infinite recursion. delay() has a custom
-// optimistic_yield-based body that intentionally avoids Arduino's __delay path.
 void delay(uint32_t ms);
 uint32_t millis();
-uint64_t millis_64();
+__attribute__((always_inline)) inline uint64_t millis_64() { return Millis64Impl::compute(millis()); }
 #elif defined(USE_LIBRETINY)
 __attribute__((always_inline)) inline void delay(uint32_t ms) { ::delay(ms); }
 // Per-variant millis() fast path — matches MillisInternal::get().