Add setBufferReleaseCallback to Speaker and Mic - #356

What

A callback that the playback / capture task calls once it has finished with a buffer given to playRaw() / playWav() / tone() or record(). From that point the caller may overwrite (Speaker) or read (Mic) the buffer.

M5.Speaker.setBufferReleaseCallback(args, [](void* args, const void* data, uint8_t channel) { ... });
M5.Mic.setBufferReleaseCallback(args, [](void* args, void* data, size_t length) { ... });

With it, two buffers used alternately are enough for runtime-generated audio. The playRaw() notes used to say three were needed; that was conservative even before (waiting for isPlaying(channel) < 2 gives the same guarantee for a single producer), and the notes now say so.

Contract (documented in the headers)

• The notice is per request, not per memory region. For playWav() the pointer is the PCM chunk inside the wav.
• Requests discarded unplayed do not call back (queued ones replaced by stop() / stop_current_sound, everything dropped by end()); the one being played when stop() cuts it does. Delivery follows the slot release, so it can arrive after isPlaying() / isRecording() already dropped: track buffers by pointer.
• Task context (never an ISR); keep it short, never block, never call begin() / end() from it. The sound may still be in the I2S queue.
• play*() / record() may be called from within: they never wait there and return false if no slot is free.
• Set or clear only before the first request or after end() returned.

Supporting changes

• play*() returns true only when the request was queued (it used to return true for a speaker that could not start and for an empty request).
• Speaker: task handle is atomic; end() is serialized with begin() under the same lock.
• _slot_occupied loads with acquire (ordering for producers that poll isPlaying()).

Examples

• Speak_with_AquesTalk / Speaker_SD_wav_file / Bluetooth_with_ESP32A2DP: two buffers with atomic busy flags.
• WebRadio / MP3 (ESP8266Audio wrappers): three buffers in rotation with the callback (the third keeps one buffer of slack against decoder / network jitter). Registered from an explicit setup() after M5.begin().
• Basic/Microphone: draws the block the callback reports as filled.

Fixes found while verifying the examples on hardware (same commit):

• Bluetooth: the phone's AVRCP volume now drives the speaker and local button changes are reported back; builds on the current ESP32-A2DP (1.8.x) as well as 1.7.x.
• WebRadio: WiFi modem sleep off (with it on, the stream buffer drained at ~2 KB/s while the link could deliver far more), 32 KB stream buffer with pre-buffering and refill, stations that cannot be opened or stop early are skipped, the station list is refreshed (the first entry returned 404), WIFI_SSID / WIFI_PASS can be passed from the build.
• Speaker_SD_wav_file plays every *.wav in the root of the card; MP3_with_ESP8266Audio plays every *.mp3 in /mp3.

Verification

• ESP32-S3 (StickS3): two-buffer ping-pong driven from the callback, and refilling from inside the callback, 0 overlaps with the task reading the buffer (checked by inspecting the task state during refills); three-buffer and isPlaying-wait forms unchanged. Basic/Microphone record / playback on hardware.
• CoreS3 SE: Speaker_SD_wav_file, MP3_with_ESP8266Audio (ESP8266Audio 2.4.1), WebRadio_with_ESP8266Audio (long play without dropouts, station switching).
• Core Basic: Bluetooth_with_ESP32A2DP with ESP32-A2DP 1.7.x on Arduino core 2.x and 1.8.11 on core 3.x, including volume from the phone.
• Not verified: Speak_with_AquesTalk (licensed library not available here); compiles with the same wrapper pattern as the others but was not built.

Breaking changes

• playRaw() / playWav() / tone() now return false when the speaker could not be started or the request is empty (previously true).

Fork CI (Arduino / ESP-IDF builds) passed on ainyan03/M5Unified before this PR.