fix(search): guard Searcher index against concurrent rebuild double-close

[walcz-de]

Problem

Searcher.IndexSkills closes and recreates the bleve index without any synchronization:

func (s *Searcher) IndexSkills(skills []Skill) error {
    s.index.Close()            // <- no lock
    os.RemoveAll(s.indexPath)
    index, err := bleve.New(s.indexPath, mapping)
    s.index = index
    ...
}

GitSyncer triggers a re-index via the onUpdate callback (RebuildIndex → IndexSkills), and re-index is driven per configured repo. When two rebuilds overlap (e.g. multiple git repos syncing, or a periodic sync racing a manual one), the second s.index.Close() closes an already-closed index and the process panics:

panic: close of closed channel
    .../pkg/git/syncer.go:177   (onUpdate callback)

This is not just an internal error — it crashes the host process. We hit it downstream in LocalAI, where the periodic skill git-sync took down the whole LLM server.

Search has the mirror race: it reads/uses s.index while IndexSkills swaps and closes it → search failed: index is closed.

Fix

Add a sync.RWMutex to Searcher:

• IndexSkills and Close take the write lock.
• Search takes the read lock, held across index.Search so a concurrent rebuild cannot Close the index mid-search.

Minimal and surgical — no behavioural change on the single-threaded path.

Test

Adds a -race regression test (pkg/domain/search_test.go) that runs 8 concurrent IndexSkills calls plus interleaved searches. It panics with "close of closed channel" without the fix and passes with it (verified under go test -race).