Recover word prediction from one slow call without needing Retry

[enaboapps]

Closes #924

Problem

Word prediction turned itself off after a single slow moment and stayed off until the person selected Retry predictions. One call over 1.5 seconds in the worker, or one reply over two seconds in the main process, was enough.

Change

• Worker: a slow call still returns its suggestions. Three slow calls in a row mark the model unavailable. A call at normal speed clears the count. A model error is still fatal at once.
• Main process: the first missed reply deadline kills the worker and its text context, and a new worker starts on the next poll. A second missed deadline within 60 seconds of that replacement fails prediction and offers Retry predictions as before, whatever the replacement answered in between. A third missed deadline while the same keyboard is open also fails.
• An acceptance waiting on the missed reply still fails and types nothing.

No protocol interface or persisted schema changes.

Validation

All run locally on Windows and passing:

• npm run prediction-model
• npm run lint
• npm test
• npm run build
• cargo fmt --manifest-path src-tauri/Cargo.toml --check
• cargo clippy --manifest-path src-tauri/Cargo.toml --all-targets -- -D warnings
• cargo test --manifest-path src-tauri/Cargo.toml: 574 passed

New tests use fake models and sleeping subprocesses and inject no input.

Not verified: behaviour in the running app under a real stall.

🤖 Generated with Claude Code