perf(audio): avoid idle buffer conversion

[JustYannicc]

Summary

• skip AVAudioConverter work on warm idle audio taps unless a meter update is due
• compute idle meter levels directly from the input buffer, while keeping recording conversion/output unchanged
• keep warmup and command behavior unchanged; conversion still happens for recording buffers and recording meters

Evidence

Focused temporary instrumentation around processBuffer counted converted, recorded, and metered buffers over matching 1.2s windows:

Scenario	Before	After
Warm idle conversions	12	0
Warm idle recorded buffers	0	0
Warm idle meter updates	7	7
Recording conversion delta	12	12
Recording append delta	12	12
Recording meter delta	8	8

This removes idle converter work while preserving recording conversion and meter cadence in the measured run.

Checks

• mcp__lsp.diagnostics on src/native/audio-capturer.swift: no errors
• swiftc -O -o /tmp/supercmd-audio-capturer-final src/native/audio-capturer.swift -framework AVFoundation -framework Foundation: passed
• git diff --check: passed
• node scripts/check-i18n.mjs: completed; existing it locale missing keys were reported and are unrelated to this native-only change
• npm run build:native: blocked by local policy; npm is disabled on this Mac, pnpm attempted install, then failed with ERR_PNPM_IGNORED_BUILDS

Risks

• Idle meter levels are now calculated from the input buffer instead of the converted 16 kHz mono buffer. The code reads channel 0 with the same RMS/peak scaling, matching the converter's channel map for typical mic formats.

[JustYannicc]

Superseded by consolidated upstream PR #534 (perf(mac): reduce native polling and async IO stalls), which carries this change together with the related macOS/native/audio/camera async IO performance fixes. Closing this source PR so review can continue on #534.