chore: sync sdk code with DeepLearning repo

[ccampbell-aai]

Syncs the latest changes from the internal repo. Version bump 0.64.29 → 0.64.31.

0.64.31

• Sync model: universal-3-5-pro is now the sole synchronous speech model and the default for SyncTranscriber, matching the Node SDK. The former u3-sync-pro name was a branding alias for the same model and has been removed from SyncSpeechModel. (Passing the raw string "u3-sync-pro" still works server-side, but SyncSpeechModel.u3_sync_pro no longer exists.)

0.64.30

• docs: Add a "Sync STT Transcription Examples" section to the README with runnable SyncTranscriber snippets — basic transcription, configuration (prompt / keyterms_prompt / conversation_context / raw PCM), language selection via language_codes, opt-in word timestamps, connection pre-warming, and error handling.
• docs: Clarify the SyncTranscriptionConfig.language_codes docstring — it defaults to None and overrides the default prompt when set (ignored when a custom prompt is provided).

Merging this PR publishes 0.64.31 to PyPI.