Improved turn handling, TTS batching and new providers

What's Changed

• feat(tts): add Fish Audio S2 local provider via vLLM-Omni by @darkliquid in #3
• Make speech synthesis batched, multi-speaker and offline-capable by @darkliquid in #4
• docs(providers): add dedicated guides for local self-hosted engines by @darkliquid in #5
• feat(provider): add Cartesia TTS (Sonic) and STT (Ink Whisper) integration by @darkliquid in #7
• Add Inworld AI as an LLM, TTS and STT provider by @darkliquid in #6
• feat(engine): replace submit_turn with a progressive turn stream by @darkliquid in #8

Full Changelog: v0.1.0...v0.2.0