feat(smallestai): use Waves continuous WebSocket streaming protocol for TTS

[harshitajain165]

Summary

The SmallestAI SynthesizeStream currently buffers all incoming tokens locally and sends one isolated request per segment, without using the Waves continuous-streaming protocol (continue / flush / max_buffer_flush_ms) that comparable LiveKit TTS plugins (ElevenLabs, Cartesia, Rime, Neuphonic, Inworld, …) use.

This PR switches the plugin to the continuous protocol so it speaks the Waves streaming API the way the official SDK and other providers' plugins do:

• Forwards tokens to /waves/v1/tts/live as they arrive with "continue": true, instead of buffering the whole segment locally.
• Closes each segment with an explicit "flush": true frame.
• Exposes the server-side max_buffer_flush_ms buffer bound as a TTS(...) constructor arg and update_options(...) option (default 0).
• Runs send/receive as concurrent tasks over the pooled connection. The connection is only returned to the pool after a clean complete; the pool discards it on error/interruption, so reuse stays safe.
• Collapses to one SynthesizeStream instance = one segment, matching the current base-class contract (multi-segment instances are deprecated upstream).

Response-frame handling (chunk / word_timestamp / complete / error) is unchanged, and the HTTP ChunkedStream path is untouched.

Scope / note

This change is about protocol parity — using the continuous-streaming fields the Waves WS API exposes, matching the official SDK and other plugins. It is not a latency change: with current server behavior the endpoint synthesizes on the explicit flush, so output timing is unchanged from the previous buffer-until-flush implementation. The client is now ready to benefit from any future server-side incremental emission with no further plugin changes.

Testing

Verified against the live Waves /tts/live endpoint:

• Continuous streaming: tokens streamed with continue/flush/max_buffer_flush_ms are accepted and return audio (lightning_v3.1 / magnus).
• Word timestamps over streaming: with the correct voice/model pairing (lightning_v3.1_pro / meher, word_timestamps=True), the endpoint returns word_timestamp frames — all words surfaced as TimedString transcript entries with sensible timings alongside audio. The word_timestamps flag on the streamed frames is accepted (no "Invalid input data").
• Interruption / reuse: cancelling a stream mid-synthesis leaves the pool clean — a subsequent stream on the same TTS instance completes normally.
• HTTP path: synthesize() (ChunkedStream) is unchanged and still returns audio.
• Hermetic tests (local mock WS server) cover the wire protocol and the word_timestamps flag — kept local for now, not included in this PR.
• ruff format, ruff check, mypy all clean.