Summary

Adds a mode streaming parameter to the AssemblyAI STT plugin, exposing the u3-pro accuracy/latency preset:

• min_latency — fastest time-to-text
• balanced — the server default, recommended for voice agents
• max_accuracy — highest accuracy, for scribes/post-call

mode is forwarded to the u3-pro ASR server, which applies its own per-mode tuning. Any silence / partials / VAD knobs set explicitly still take precedence over the mode's defaults.

Details

• u3-rt-pro-family only (u3-rt-pro, u3-rt-pro-beta-1, universal-3-5-pro) — passing mode with any other model raises ValueError, consistent with the other u3-pro-gated params (prompt, voice_focus, etc.).
• Connect-time only, not exposed via update_options. This mirrors the official AssemblyAI Python SDK, where mode lives on the connect-time StreamingParameters and is not part of UpdateConfiguration.
• Sent as a connection query parameter; omitted entirely when unset so the server default applies.

Implementation follows the existing voice_focus param precedent (#6119).

Test plan

New tests in tests/test_plugin_assemblyai_stt.py:

• default (unset)
• accepts each of the three values
• raises for non-u3-pro-family models
• accepted for every u3-pro-family model
• present in the connect config query when set / absent when unset
• not exposed via update_options (connect-time only)

All 58 plugin tests pass; ruff format, ruff check, and mypy clean.