feat(models): upgrade MAI-Transcribe-2 and Muse Spark 1.3

[MyButtermilk]

Summary

• replace the Azure default with the exact MAI-Transcribe-2 model and use microsoft/mai-transcribe-2 for new OpenRouter transcription work
• replace Muse Spark 1.2 Standard/Contributor with their corresponding 1.3 tiers
• migrate persisted settings and Meeting analysis selections without changing custom model codes or historical analysis results
• retain the exact MAI 1.5 contracts for explicit Azure overrides and already frozen jobs, including recovery of durable provider results without a second upload
• integrate current main (70697f0) and preserve PR refactor: remove dead code, redundant wrappers, and low-value tests #41's code/test cleanup, including the removal of start.sh

Review fixes

• Restore legacy Azure/OpenRouter capability IDs and revisions; keep Azure's MP3 preparation.
• Allow only the exact current and legacy OpenRouter models at both direct transport and buffered pipeline boundaries. New admissions still select version 2.
• Migrate stored meetings.analysis_model values at database initialization; preserve analysis history, timestamps, and cached model provenance.

Integration details

Azure retains its LLM Speech endpoint and API version 2025-10-15; version 2 supports the existing phraseList path and southeastasia is accepted. UI metadata reflects the published limited-time $0.10/audio-hour price through 2026. Muse retains the existing Meta transport, token cap, tier pricing, and Contributor data-use warning. MAI routes continue to expose final text only.

Validation

• 876 local Python tests passed across settings/API/summarization, Meeting persistence, MAI transport/capabilities, pipeline routing, and queued/paid-result recovery.
• Repository-wide Ruff lint and formatting passed.
• Frontend application/test typechecks and ESLint passed; 109 library tests and 63 component tests passed.
• Independent review of the fixes and integration with current main found no blockers.
• All six PR CI jobs passed on the updated head, including the full Windows Python suite, real-browser File smoke, extended mypy, and Rust fmt/clippy/tests: https://github.com/MyButtermilk/Scriber/actions/runs/33949371126
• Codex re-reviewed f611da4127 and reported no major issues; all original review threads are resolved.
• Local checks used Python 3.14.6 and Node 24; Windows CI passed with pinned Python 3.14.7 / Node 26.5.
• Two unchanged device-test cases initially exposed an existing database-initialization dependence; both isolated tests and the identical 561-test combined rerun passed. Their production paths and test bodies are unchanged from main.
• Provider transport validation uses fixtures/mocks; no live provider API request was made.

References

• https://learn.microsoft.com/en-us/azure/ai-services/speech-service/mai-transcribe
• https://learn.microsoft.com/en-us/azure/ai-services/speech-service/regions?tabs=llmspeech
• https://microsoft.ai/models/mai-transcribe-2/
• https://openrouter.ai/microsoft/mai-transcribe-2
• https://ai.developer.meta.com/docs/models/
• https://research.meta.ai/blog/introducing-muse-spark-1-3

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 95d12e2418

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[MyButtermilk]

@codex review
Review head: f611da4127ab5c0fa057b0e63abd39a4f3e5f8f7

Please focus on the three addressed findings: MAI 1.5 frozen-route and durable-result recovery, explicit Azure overrides, and durable Muse Meeting model migration. Also check that integration with current main preserves PR #41's cleanup. All three original threads now include the fix and regression evidence.

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Nice work!

Reviewed commit: f611da4127

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".