Use AssemblyAI's latest default speech model

[gthay]

Summary

• remove the explicit universal-3-pro / universal-2 model list from AssemblyAI transcript requests
• let AssemblyAI automatically select its latest default speech recognition model
• add a regression test that verifies transcript payloads omit speech_models

Why

AssemblyAI will deprecate universal-3-pro on September 2, 2026, after which requests that explicitly name it will fail. The backend currently pins that model in every transcript request.

Because speech_models is optional, omitting it lets AssemblyAI route requests to its current Universal default. This avoids the immediate deprecation failure and future app changes solely for renamed default models. See AssemblyAI's model selection documentation.

Impact

AssemblyAI cloud transcription continues working after the deprecation and automatically follows AssemblyAI's latest default model. Existing API keys, language hints, and custom spelling behavior are unchanged.

Validation

• git diff --check
• verified production sources contain neither the deprecated model ID nor an emitted "speech_models" payload key
• added testAssemblyAITranscriptRequestUsesLatestDefaultModel
• swift test could not run in the local Linux workspace because the Swift toolchain is unavailable; the repository's macOS CI will run the full builds and test suite

[gthay]

Hi @yazinsai,
please review this PR carefully, it's generated with Sol 5.6 and merge.

[yazins-ai]

Thanks @gthay for updating the AssemblyAI integration to use their latest default speech model — merged. 🎉