[codex] Improve model download progress

[IchenDEV]

Summary

Improves model download feedback and makes speech model downloads explicitly user initiated.

What Changed

• Added shared download progress metrics for elapsed time, percentage, remaining bytes, and current speed.
• Shows the same download detail in the menu bar, model management rows, and onboarding.
• Stops automatic speech model downloads when switching ASR type or attempting to use an undownloaded speech model.
• Shows a visible error HUD when the current speech model has not been downloaded.
• Updates English and Simplified Chinese copy to clarify that speech models must be downloaded manually from Settings > Models.

Why

Previously, some speech model paths could start downloads implicitly, and users did not get a clear visible error when trying voice input with an unavailable model. This makes the behavior predictable: downloads only happen from the Download button, and missing models are reported immediately.

Validation

• git diff --check
• swift test (115 passed, 3 skipped, 0 failures)