fix: transcript fallback only reads last assistant message

fix: installer dedup normalizes ~ vs absolute paths to prevent duplicate hooks

fix: forked sessions no longer replay entire transcript TTS

fix: default voice_cue_mode to off

fix: default messaging_updates to off

fix: use pythonw.exe + CREATE_NEW_PROCESS_GROUP for Windows bg server

fix: Windows bg server uses shell redirect for proper detach

fix: use DETACHED_PROCESS for Windows background server launch

fix: lazy-import TTSEngine to avoid loading torch for piper engine

fix: background server fd closure crash on Windows

fix: start command no longer overrides .env voice config

fix: use platform-aware config path on Windows for server .env (#9)

chore: bump version to 0.2.7

fix(copilot): extract mood from tts tags, buffer partial lines, guard ghc payload

feat(copilot): events.jsonl watcher replaces script-based TTS capture