feat: optimize skill discovery and runtime synchronization

[alcholiclg]

Change Summary

• Add metadata-only skill discovery with marker-based caching while preserving existing bounded scan semantics.
• Generate support-file signatures during the existing materialization pass and add an opt-in no-I/O runtime synchronization path; existing callers retain full synchronization by default.
• On the 1,226-file OneSkills fixture, discovery p95 drops from 56.957 ms to 0.236 ms and unchanged runtime - synchronization from 84.998 ms to 0.000167 ms.
• Add discovery, cache invalidation, scan compatibility, signature, and runtime synchronization tests.

Related issue number

Checklist

• The pull request title is a good summary of the changes - it will be used in the changelog
• Unit tests for the changes exist
• Run pre-commit install and pre-commit run --all-files before git commit, and passed lint check.
• Documentation reflects the changes where applicable