Zerm 2.8.2: rebuild Meetings and unify Speech workflows

[thefourcraft]

Summary

• rebuild Meetings recording around an application-scoped lifecycle, independent timestamped room/call tracks, bounded real-time handoffs, interruption recovery, durable manifests, and synchronized review playback
• use the selected Dictation model/provider/language for meeting transcription, with local canonical processing, cloud coverage retries, provider-aware language behavior, speaker confidence, and local Ollama summary preflight
• repair short Hebrew Dictation auto-detection with Whisper language-aware candidates, VAD recovery, and a guarded Hebrew comparison pass
• rebuild local Read Aloud so the selected on-device LLM analyzes the complete selection, preserves Hebrew, and then streams synthesized speech; add Retell/Summarize/Explain/Simplify, local history, and optional Apple system voices
• make Instant + Refine the Enhancement default and require enhancement output to preserve the source language, including Hebrew
• reorganize navigation into expandable Dictation, Meetings, and Read Aloud workspaces, with Enhancement as a global Speech destination and native global Settings
• enforce meeting-safe Read Aloud routing: headphones allowed, speakers blocked, disconnect stops TTS and notifies while Dictation/meeting capture continue
• add Hebrew localization, permission strings, accessibility semantics, deterministic UI-test hosting, security/privacy hardening, CI/release safeguards, and current Notebook documentation
• stamp version 2.8.2 / build 282

Office Mac verification

• exact latest source fingerprint matched the feature worktree before validation
• current unit suite: 136/136 passed across 23 suites, zero failures/skips (/tmp/Zerm-hebrew-readaloud-final-units.xcresult)
• current Release build: succeeded, arm64 2.8.2 (282), zero compiler warnings/errors (/tmp/Zerm-hebrew-readaloud-final-release2.log)
• opt-in real integration: installed Hebrew Apple voice rendered speech and Whisper Auto returned predominantly Hebrew
• opt-in real integration: installed Gemma generated a natural Hebrew retell and Hebrew Enhancement output
• opt-in real integration: installed Kokoro synthesized and streamed audio through AVAudioEngine to completion
• current UI target compiles; the updated six-test runtime suite is pending an unlocked Office Aqua session. The previous exact 2.8.2 tree passed 6/6 before this Read Aloud/IA follow-up.
• no app build, test, or launch was performed on the primary development Mac

Release status

This PR intentionally remains draft. The current internal Release artifact is ad-hoc because the Office keychain has no signing identity; the existing /Applications/Zerm.app remains Developer-ID signed and was not overwritten. Current UI execution, Developer ID signing, notarized DMG/ZIP, Office install, Gatekeeper check, and Sparkle 281 to 282 update proof remain required before merge/publication.

Hardware follow-up matrix

Office automation has no microphone. Before publication, validate real microphone and conferencing-app/system-audio capture; natural English/Hebrew speech and multi-speaker diarization; wired/AirPods-Bluetooth/USB/analog routing and disconnect behavior; VoiceOver, contrast, and broader RTL layouts.