refactor: TTS provider based on unified SDK design (#48)

* refactor: tts provider based on unified sdk design

- Add unified public types (TextInput, VoiceSelection, AudioConfig,
  SpeechParams, SynthesizeRequest, etc.)
- Define core interfaces (Synthesizer, StreamingSynthesizer,
  SynthesisStream, WarmableProvider)
- Add Capability types (Capabilities, ModelCapability, Feature) for
  provider query
- Add parameter normalization helpers (speed/pitch/volume/emotion
  mapping)
- Simplify Config to APIKey/Endpoint/Model/Voice/SampleRate/Extra
- Add ProviderMeta.Description, Features; ModelInfo.SystemVoices;
  VoiceInfo.Emotions
- Provider = Synthesizer, StreamingProvider = StreamingSynthesizer type
  aliases
- Move DashScope adapter to new interfaces, read provider-specific
  params from Extra
- Add complete cosyvoice-v3-flash voice list (50 voices) in meta.go
- Update TTSProcessor to use SynthesizeRequest internally
- Simplify config.TTSConfig, voicebot.example.json

* Sync system providers/models/voices to database via content hashes

Add deterministic SHA-256 hashing to ASR/TTS/LLM provider metadata and
implement `SyncSystemProviders()` that incrementally syncs
code-registered system resources with the database,
auto-creating/updating records whose hash changed and removing stale
entries.

* Switch voice and model listing to language-prefix matching

Use the `language.Match` helper for prefix-based language filtering
instead of exact string equality. Apply this consistently to both
system voices and speech models in the available list endpoint.

* Remove deprecated fun-asr model from Aliyun Dashscope ASR

---------

Co-authored-by: liuscraft <liuscraft@users.noreply.github.com>

feat(manager): align dashboard and API docs (#41)

feat: 增加生产部署与服务健康检查 (#37)

* feat: add production deployment stack

* ci: publish images on release