[ZhuoenX]

The public audio wrappers supply all optional arguments to the native extension, but the spectrogram, mel_spectrogram, and melscale_fbanks parsing formats do not match those arguments. Correct the formats and parse boolean arguments into int temporaries before assigning bool fields, preventing adjacent-field corruption. Also fix hanning_window to call the registered native hann_window function.

Document the public Python API, its keyword support, real window constants, WAV I/O, and current spectral-feature behavior. Add a runnable WAV feature-extraction demo with optional output saving and explicit failures.

Validation on macOS arm64/Python 3.9.6:

• Eight compiled parser regression checks covering defaults, all arguments, and independent boolean fields.
• Rebuilt MNN and the actual _mnncengine extension from this branch; ran WAV loading/saving, spectrogram, mel, fbank, and Whisper features successfully.
• Two numerical checks covering center padding, normalization, and nondefault mel options against an explicit filter-bank projection.
• Python 3.8 grammar, local documentation links, git diff whitespace, and changed-line C++ formatting checks pass.

Older packages with the affected bindings need a build containing these fixes. The existing sampe_rate keyword remains unchanged for compatibility.

[CLAassistant]

All committers have signed the CLA.