chore: sync sdk code with DeepLearning repo

[bgotthold-aai]

• Rename noise_suppression_* → voice_focus_*. New canonical wire keys: voice_focus, voice_focus_threshold. -
• Add continuous_partials — bool, U3-Pro steady-partials mode.
• Add customer_support_audio_capture — bool. Emits a runtime warning when enabled since it captures session audio to AssemblyAI's support bucket.
• Add Word.speaker for the new word-level speaker labels emitted on each word when speaker_labels=true.