All code comments, docstrings, and log messages translated to English.
Chinese strings are preserved ONLY where they are legitimate i18n data:
Kept (user-facing language data):
- self.system_prompt: LLM fine-tuning system prompt (must match training data)
- initial_prompt: Whisper domain hint vocabulary
- _fix_recognition.replacements: homophone correction table
- All regex patterns in _try_simple_parse / _fallback_parse (they match
actual spoken Chinese commands — this IS the i18n layer)
Translated to English:
- All print() log messages
- All inline comments and docstrings
- Class/section header comments
Part of #5. No functional changes.
