[codex] Add audio storage retention controls

[r3dbars]

Summary

• Add audio retention settings so transcripts stay durable while retained meeting audio can be kept, compressed, or aged out.
• Backfill existing meeting WAV archives into owner-only M4A files and clean stale conversion temps.
• Harden destructive cleanup so it only touches Transcripted-owned meeting audio, skips symlinks, preserves unrelated/forged/orphaned audio, and keeps failed-meeting retry audio safe.

Why

Users can build up huge uncompressed WAV archives. The app should stop quietly eating disk while still preserving transcripts and safe replay audio by default.

Verification

• bash run-tests.sh — 1413 passed
• swift test — 119 passed
• bash run-integration-smoke.sh — passed
• bash build-deps.sh --force — passed
• bash build.sh — signed build passed
• Controlled old-user migration launch — passed: fresh WAV converted to valid owner-only M4A, old retained audio pruned while Markdown stayed, stale temps removed, user/forged/failed audio preserved
• Real library scan — 0 WAVs, 66 M4As, 0 hidden temp M4As, 0 symlinked audio dirs, 40M retained audio

Release notes

This should not ship until CI is green and the normal release path is run: notarized package, Sparkle appcast, Homebrew cask, and website/download verification.

[r3dbars]

Packaging smoke also passed after opening the PR.

• SKIP_NOTARIZATION=1 bash build-beta.sh transcripted-release-smoke Codex passed
• Developer ID app signing verified with hardened runtime
• DMG assembled and signed: build/Transcripted-1.1.30.dmg (504M)
• Notarization was intentionally skipped for this dry run, so Gatekeeper rejection is still expected until the real release notarization/stapling pass

GitHub reports the PR is mergeable, and no CI checks are configured/reported for this branch.