fix: correct sample rate detection for Bluetooth devices

[pasrom]

Summary

• Fix wrong-scope kAudioStreamPropertyPhysicalFormat query that returned BT HFP rate (24kHz) instead of tap delivery rate (48kHz) on AirPods
• Add kAudioTapPropertyFormat query on tap directly — authoritative source for actual data rate
• Add kAudioDevicePropertyActualSampleRate verification at first IOProc callback
• Add post-hoc mic-duration cross-check as safety net in DualSourceRecorder.stop()
• On nominal/stream rate mismatch, prefer nominal over stream (stream scope reflects BT HFP, not tap rate)

Fixes #82

Root Cause

queryStreamSampleRate queried kAudioStreamPropertyPhysicalFormat with kAudioObjectPropertyScopeOutput on the aggregate device. Tap audio arrives as an input stream — output scope returns the physical output device's format, which on BT HFP/SCO drops to 24kHz. The cross-validation then incorrectly preferred this 24kHz, causing AudioMixer.resample(from: 24000, to: 16000) on data that was actually at 48kHz → 2x speedup.

Fix (3 layers)

1. Primary: kAudioTapPropertyFormat on tap object — same approach as AudioCap and Recap
2. Secondary: kAudioDevicePropertyActualSampleRate at first IOProc callback — hardware-measured rate
3. Tertiary: Post-hoc cross-check comparing raw file size to mic WAV duration, snapped to nearest standard rate

Test plan

• 12 new unit tests for snapToStandardRate and inferRateFromDuration (SampleRateQuery)
• 5 new unit tests for crossCheckAppRate (DualSourceRecorder)
• All 31 AudioTapLib tests pass
• All 800 app tests pass
• All 53 E2E/integration tests pass
• Manual test with meeting-simulator: detection → recording → transcription → protocol (correct duration, no Mickey Mouse)
• Lint clean