feat: Add match_original_utterance translation, clean speaker identification schema, update streaming language detection/VAD threshold

[gsharp-aai]

API Reference (llm-gateway.yml & openapi.yml):

• Added match_original_utterance parameter for translation (returns translated text per utterance when Speaker Labels enabled)
• Added translated_texts field to TranscriptUtterance schema
• Added name_role to speaker_type enum
• Updated known_values description (now optional for all speaker types, not just "role")
• Removed non-existent format_utterances parameter from custom formatting
• Updated custom formatting response schema (made formatted_text and formatted_utterances optional)
• Added complete response example showing all three speech understanding features together

Streaming Changes (usm-streaming.yml):

• Enhanced language_detection description to clarify it returns metadata on utterances and final turns
• Added vad_threshold parameter (0.0-1.0) for VAD confidence tuning in noisy environments

Documentation Changes:

• Removed format_turns dynamic parameter setting code from streaming examples (not necessary/confusing)
• Updated multilingual language detection description to mention final turns

[github-actions]

🌿 Preview your docs: https://assemblyai-preview-33cb0020-257c-456b-ae55-7f8056050d1e.docs.buildwithfern.com/docs

[github-actions]

🌿 Preview your docs: https://assemblyai-preview-1bc49c4a-de6b-4001-8c55-a08e9069ccce.docs.buildwithfern.com/docs

[github-actions]

🌿 Preview your docs: https://assemblyai-preview-e82434ce-7a68-4a07-952d-13851b97656a.docs.buildwithfern.com/docs