[ainyan03]

Pace the Mic_FFT drawing by the captured audio

• loop() advanced one drawing step whenever two recording requests were queued, whether or not new audio had arrived. The history row scrolls once per five steps, so its speed depended on how fast the display draws. On the PC (SDL) build it scrolled hundreds of rows per second over the same data.
• A drawing cycle now starts only after a new block has been queued, so the history follows the capture rate. On a display that is slower than the capture rate, nothing changes.

Build the example on PC

• decltype instead of the GNU typeof extension, and malloc / free instead of heap_caps_* outside ESP_PLATFORM, so the example builds as is with the native (SDL) build.

Testing

• Built for ESP32-S3 (Arduino 3.x, ESP-IDF 5.5), ESP32 (Arduino 2.x) and native (SDL).
• Ran on the native build (macOS) with the PC microphone: the spectrum and history follow the input, and the history no longer races.