- Added cross-sentence stitching support to NvidiaTTSService. Multiple sentences within an LLM turn are fed into a single `SynthesizeOnline` gRPC stream, enabling Magpie's seamless audio across sentence boundaries (requires Magpie TTS model v1.7.0+).
- Added new parameters to NvidiaTTSService:
  - `custom_dictionary` for IPA-based custom pronunciation
  - `encoding` for output audio encoding format
  - `zero_shot_audio_prompt_file` for zero-shot voice cloning with file caching
  - `audio_prompt_encoding` for zero-shot audio prompt format
- Added `can_generate_metrics` returning True for NvidiaTTSService.
- Added `stop_all_metrics()` call on audio context interruption.
- Added gRPC error handling for synthesis config retrieval.