Add fuzz tests

[Vladyslav-Kuksiuk]

This PR:

• Adds fuzz tests.
• Updates skills to run linter for verifications.
• Hardens pattern matching against gobwas/glob panics so bad matcher inputs report normal pattern-not-found errors instead of crashing.

Resolves this issue.

[Oleg-Melnik]

Claude Review

Observation — matchGlob recovers per line × per matcher inside the scan hot path

In embedding/parsing/pattern.go, matches calls matchGlob, which sets up a defer/recover closure on every invocation. Since FindIn → matchesAt → matches runs once per (candidate start line × matcher), a large source file scanned per instruction pays a defer + closure setup for every line.

This is a deliberate design choice rather than a bug: recovering per line means a panic on a single line is treated as a local miss, so scanning can still find a match at a later start position. Wrapping the whole FindIn loop in a single recover instead would be cheaper but would abort the entire scan on any panic, losing that precision.

Just flagging the trade-off for visibility — the per-line granularity costs some hot-path overhead on large files in exchange for match precision. No change required if that trade-off is intentional.

[Oleg-Melnik]

@Vladyslav-Kuksiuk LGTM with a comment to address before merging.