fix(#388): address PR review findings — fix WebSocket/REST bugs, improve error handling, fix types and comments

Critical fixes:
- Fix FormData field name mismatch (audio -> file) to match backend FileInterceptor
- Add /speech namespace to WebSocket connection URL
- Pass auth token in WebSocket handshake options
- Wrap audio.play() in try-catch for NotAllowedError and DOMException handling
- Replace bare catch block with named error parameter and descriptive message
- Add connect_error and disconnect event handlers to WebSocket
- Update JSDoc to accurately describe batch transcription (not real-time partial)

Important fixes:
- Emit transcription-error before disconnect in gateway auth failures
- Capture MediaRecorder error details and clean up media tracks on error
- Change TtsDefaultConfig.format type from string to AudioFormat
- Define canonical SPEECH_TIERS and AUDIO_FORMATS arrays as single source of truth
- Fix voice count from 54 to 53 in provider, AGENTS.md, and docs
- Fix inaccurate comments (Piper formats, tier prop, SpeachesProvider, TextValidationPipe)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

docs/SPEECH.md

@@ -494,7 +494,7 @@ Boolean parsing: `value === "true"` or `value === "1"`. Unset or empty values de
 
 **Capabilities:**
 
-- 54 built-in voices across 8 languages
+- 53 built-in voices across 8 languages
 - Speed control: 0.25x to 4.0x
 - Output formats: mp3, wav, opus, flac
 - Voice metadata derived from ID prefix (language, gender, accent)