Add AudioEmulator for microphone/getUserMedia testing

[noahgift]

Problem

Testing real-time audio streaming applications (like whisper.apr) requires mocking navigator.mediaDevices.getUserMedia() to inject controlled audio. Currently this requires ad-hoc JavaScript injection:

// Current approach (demos/tests/src/browser_tests.rs:862)
let mock_media_js = r#"
    navigator.mediaDevices.getUserMedia = async function(constraints) {
        if (constraints.audio) {
            return window.__mockAudioStream;
        }
    };
"#;
page.eval_wasm(mock_media_js).await?;

This violates probar's "ABSOLUTE ZERO JavaScript" principle.

Proposed Solution

Add AudioEmulator to probar::emulation:

pub struct AudioEmulator {
    sample_rate: u32,
    channels: u32,
}

pub enum AudioSource {
    /// Inject WAV file
    WavFile(PathBuf),
    /// Generate sine wave
    SineWave { frequency: f32, duration_ms: u32 },
    /// Generate silence
    Silence { duration_ms: u32 },
    /// Generate speech-like audio (for VAD testing)
    SpeechPattern { pattern: Vec<(f32, u32)> }, // (level, duration_ms)
}

impl AudioEmulator {
    pub fn inject(&self, page: &Page, source: AudioSource) -> ProbarResult<()>;
    pub fn stop(&self, page: &Page) -> ProbarResult<()>;
}

Use Cases

1. VAD Testing: Inject audio with known speech/silence patterns to verify state transitions
2. Streaming Tests: Test chunk accumulation and transcription triggering
3. Performance Tests: Measure RTF (real-time factor) with known audio length
4. Edge Cases: Test DC offset, clipping, sample rate conversion

Related

• whisper.apr demo uses streaming transcription that needs this for proper E2E testing
• Current tests use sine wave oscillator which doesn't trigger VAD properly

Toyota Way

• Poka-Yoke: Type-safe audio injection prevents format mismatches
• Muda: Reusable component eliminates ad-hoc JS injection

[noahgift]

Triage: P3 — WASM testing feature. AudioEmulator for microphone/getUserMedia testing. Part of WASM UX testing epic (#14).