perf(audio): bulk-write captured WAV samples

[JustYannicc]

What changed

• Convert captured float samples into one contiguous little-endian Int16 PCM buffer before appending it to the WAV Data.
• Add scripts/test-audio-wav-bulk-write.swift to validate byte-for-byte legacy/bulk WAV output, header/data sizes, and the 30-second 16 kHz mono serialization timing.

Why

stopRecording() and takeSnapshot() both serialize captured WAV data on the native path. Live dictation snapshots call this periodically, so the previous two-byte Data append per sample added avoidable work on every snapshot.

Compatibility impact

• WAV output remains RIFF/WAVE PCM, mono, 16-bit little-endian samples, using the passed sample rate.
• Sample clamping/scaling and atomic file write behavior are unchanged.
• No TypeScript bridge or API shape changes.

How tested

• swiftc -O -o /tmp/supercmd-audio-capturer-check src/native/audio-capturer.swift -framework AVFoundation -framework Foundation
• swiftc -O -o /tmp/supercmd-audio-wav-bulk-test scripts/test-audio-wav-bulk-write.swift
• /tmp/supercmd-audio-wav-bulk-test
  • samples=480000 sampleRate=16000 iterations=5
  • validated byte-for-byte WAV output and header/data sizes
  • per-sample append: 15.27 ms avg checksum=617111545
  • bulk Int16 append: 0.83 ms avg checksum=617111545
  • speedup: 18.34x

Stack validation

• Clean PR branch created from origin/main at 9bdd3d2.
• Cherry-picked only this fix from the integration-stack implementation branch; PR branch contains one commit: 9d591c7.
• Draft until stack PRs perf(runtime): reduce extension and menubar churn #530, perf(ai): batch streaming and status updates #531, fix(validation): restore project checks #532, perf(renderer): cache icons and virtualize launcher surfaces #533, perf(mac): reduce native polling and async IO stalls #534, and perf(search): reduce indexing and ranking overhead #535 land.
• Stack dependency: fork/codex/perf-consolidated-integration-stack.

[JustYannicc]

Superseded by #556, which consolidates this AI/native media performance batch.