Use configurable transcribe.cpp output budget

[vladkalinichencko]

Depends on handy-computer/transcribe.cpp#74. Until that is merged and released, the Cargo patch points to the PR branch so this integration can be reviewed and tested against the exact implementation.

This keeps the Handy-side change small:

• request a 4096-token output budget for transcribe.cpp runs
• keep non-empty partial text when transcribe.cpp reports OUTPUT_TRUNCATED
• show the backend error message when history re-transcription fails
• update the integration to the current transcribe.cpp 0.2.0 API

Validated with:

• cargo fmt --check
• cargo check
• 151 Rust tests
• frontend production build
• full Rust build
• headless transcription through Handy with an installed Whisper Medium model on Metal

After #74 is released, the temporary Cargo patch entries can be replaced by the released crate version.

[cjpais]

Can you add a screenshot for this?

[vladkalinichencko]

Yep, here you go: