feat(audio): Add audio module with mel spectrogram, resampling, streaming

Implements GH-131: Comprehensive audio module for ASR/TTS applications.

## Added
- `audio::mel` - Mel spectrogram computation with MelFilterbank
- `audio::resample` - High-quality sample rate conversion
- `audio::stream` - Chunked streaming primitives
- `audio::AudioError` - Unified error type
- `audio::DecodedAudio` - Decoded audio container

## Features
- `audio` - Core audio processing (rustfft, thiserror)
- `audio-capture` - Platform audio capture (future)
- `audio-playback` - Platform audio output (future)
- `audio-codec` - Format decoding (future)

This unblocks whisper.apr CI which depends on the audio feature.

Closes #131

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: noahgift

Cargo.lock

@@ -89,6 +89,10 @@ rmp-serde = "1.3"  # MessagePack for .apr metadata (spec §2)
 rand = "0.8"
 rand_chacha = "0.3"  # ChaCha20 PRNG for Monte Carlo simulations
 
+# Audio processing
+rustfft = { version = "6.2", optional = true }  # FFT for mel spectrogram computation
+thiserror = { version = "2.0", optional = true }  # Error handling for audio module
+
 # Parallelization for graph algorithms (optional for WASM compatibility)
 rayon = { version = "1.10", optional = true }
 
@@ -141,6 +145,10 @@ format-quantize = ["half"]  # Enable quantization for .apr format (spec §6.2)
 format-homomorphic = []  # Enable homomorphic encryption for .apr format (spec: homomorphic-encryption-spec.md)
 # Note: mmap is automatic on native platforms, no feature needed (spec: bundle-mmap-spec.md)
 hf-hub-integration = ["hf-hub", "dirs"]  # Enable Hugging Face Hub integration (GH-100)
+audio = ["rustfft", "thiserror"]  # Enable audio processing (mel spectrogram, resampling)
+audio-capture = ["audio"]  # Enable audio capture (ALSA, CoreAudio, WASAPI, WebAudio)
+audio-playback = ["audio"]  # Enable audio playback
+audio-codec = ["audio"]  # Enable audio codec decoding (WAV, MP3, AAC, FLAC, Opus)
 safetensors-compare = ["safetensors", "hf-hub-integration", "half"]  # Enable SafeTensors comparison (GH-121)
 inference-monitoring = ["entrenar"]  # Enable inference monitoring and explainability
 gpu = ["trueno/gpu"]  # Enable GPU acceleration via trueno wgpu backend