fix(tui): rank skill autocomplete by bare name

[yearth]

Summary

Fix skill slash autocomplete ranking for bare queries.

Problem

Skill commands are fuzzy-matched using the full skill:<name> string. The fixed skill: prefix can influence the score more than the actual skill name. With these loaded skills:

• skill:deep-research
• skill:research-idea

typing /idea ranked skill:deep-research above skill:research-idea in a local reproduction. The expected result is skill:research-idea, because idea occurs in the bare skill name.

Changes

• Match skill commands by their bare name for non-explicit queries.
• Preserve full-command matching for explicit /skill:... queries.
• Add regression coverage for the reported ranking and both preserved query paths.

Fixes #8813

Validation

• Focused skill autocomplete tests: passed
• npm -w packages/tui test: passed
• npm run check: passed
• ./test.sh: has unrelated failures in other workspace/generated-model tests in this checkout; the packages/tui tests and checks pass.

This PR was prepared with AI assistance and reviewed by the contributor.

[yearth]

@mitsuhiko gentle ping — small TUI fix when you have a moment.

Skill autocomplete was nocase-fuzzying the full skill:<name> string, so the fixed prefix skewed scores (e.g. /idea ranked skill:deep-research above skill:research-idea). This matches non-explicit queries on the bare name while preserving full-command matching for explicit /skill:... queries, with regression coverage for both paths.

CI green, mergeable, fixes #8813. Happy to adjust anything.