Hold suggestions until the next word after context is lost mid-word

[enaboapps]

Closes #926

Problem

After the prediction context was cleared in the middle of a word, the remaining letters were completed as if they began a new word. After hel, a further l could be completed to like, leaving hellike.

Change

Suggestions are held until a character that ends the word is typed, then resume with the next word. The badge reads "Suggestions resume next word" while held.

Context is cleared by	Suggestions are held
A failed key, a failed acceptance, Retry predictions, a worker replaced after a missed deadline	When a word is in progress
A shortcut	When a word is in progress
An arrow key, Home, End, Page Up, Page Down, Delete or another untyped key	Always
Backspace that deletes text the worker does not have	When the character before the caret is part of a word or unknown
Enter or Tab	Never
A foreground change or outside keyboard and mouse activity	Never, and these end a hold

• Backspace releases a hold once it deletes back to a space or punctuation Switchify typed.
• The main process remembers, for each character Switchify typed, only whether it was part of a word. It stores no text, and at most 512 entries.

No pipe message, protocol interface or persisted schema changes.

Behaviour changes to be aware of

• After a navigation key, the next word gets no suggestions.
• Backspacing text that was there before the keyboard opened, then typing, gets no suggestions until the next word.

Both were cases where the old behaviour could complete a fragment into the wrong word. A hold costs at most one word of suggestions and never changes what is typed.

Difference from the issue

The issue proposed holding after any reset. Window changes and outside activity are excluded, because the caret may then be anywhere and typing usually starts a new word.

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: 583 passed

New tests call the service directly with fake strokes and inject no input.

Not covered by a test: the failed acceptance path, which sits inside poll and needs an app handle. Not verified: behaviour in the running app.

🤖 Generated with Claude Code