Implement a word-level scoring engine for the command palette that replaces
the previous FZF-based approach. Query tokens are matched against pre-tokenized
words in each column (key, action, category) with exact, prefix, and
edit-distance scoring. Multiple search terms are supported with cross-column
matching — items matching more tokens rank higher. Compound query tokens
containing delimiters (e.g. mouse_selection) are matched as units.
Add comprehensive tests using a Go builder API instead of raw JSON blobs,
covering single-token, multi-token, partial-match, ranking, mouse binding,
and unmapped action scenarios.
Add documentation for the command palette kitten.
6a3fe6ae01