The TUI has had instant fuzzy search since v0.1. The CLI has been blind: once a project has >20 tasks, finding one means scrolling ls output manually.

This adds the same search affordance to scripts and one-shots.

Examples

$ tsk search milk
[ ] #3 buy milk  (medium)

$ tsk search "fix login" --limit 5
[ ] #12 fix login bug  (high)
[ ] #7 add login telemetry  (medium)

$ tsk search dev --all --format table
ID  DONE  P  DUE  TITLE                  TAGS
#1  [x]   H       ship table format PR   #dev
#5  [ ]   M       review v0.3.0 release  #dev

$ tsk search docs --json | jq '.[].Title'

Implementation

• Reuses internal/util/fuzzy (sahilm/fuzzy) — same scoring the TUI uses, so CLI and TUI rank results identically
• Default index: title + #tag1 #tag2 + notes
• --title-only restricts matching to titles
• Filter: undone by default; --done for done only; --all for both
• --limit N caps results
• --format plain | table | json (table reuses the renderer from feat(ls): add --format=table with aligned columns #6)
• --json shortcut preserved, mutually exclusive with --format
• Empty result: no matches for <q> in text mode, [] in JSON mode

Tests

10 cases in search_test.go:

1. Title match
2. Tag match (default index)
3. --title-only skips tags
4. Done filtered out by default
5. --all includes done
6. --limit caps results
7. --json non-empty
8. --json empty array
9. Whitespace-only query rejected
10. --format=table works

All green with -race.

Stacked on #6

This PR builds on feat/ls-table (PR #6) for the table renderer. Once #6 merges, this PR auto-rebases against main cleanly.

Part 3/10 of the v0.3.0 improvement series.