feat: persist voice profiles to Voice_Profile/ with auto-load on startup

perf: add TTS latency optimizations