fix(opencode): stabilize duplicate skill discovery

[ualtinok]

Issue for this PR

Closes #32202

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

Makes duplicate skill discovery deterministic.

The skill catalog already renders sorted by skill name, but duplicate skill names were resolved while loading files with unbounded concurrency. Since duplicate names overwrite the previous entry, whichever file finished loading last won. That made the selected skill location unstable across restarts when the same skill name existed in multiple roots.

This PR sorts matches within each scan and loads discovered skill files in discovery order, preserving the existing source precedence while removing the I/O race.

How did you verify your code works?

• bun test test/skill/skill.test.ts
• bun typecheck

Screenshots / recordings

N/A

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

Potential duplicate found:

• PR fix(skill): deduplicate by basename during discovery to fix non-deterministic enumeration #30184 - fix(skill): deduplicate by basename during discovery to fix non-deterministic enumeration
  • fix(skill): deduplicate by basename during discovery to fix non-deterministic enumeration #30184
  • Reason: This PR is directly related to your current PR fix(opencode): stabilize duplicate skill discovery #32203. Both address non-deterministic skill discovery behavior. PR fix(skill): deduplicate by basename during discovery to fix non-deterministic enumeration #30184 handles deduplication by basename, while your PR fix(opencode): stabilize duplicate skill discovery #32203 stabilizes duplicate skill discovery by sorting matches and loading files in discovery order. These appear to be addressing the same or closely related stability issues with skill enumeration.