Fix #537: protect rutor state with RWMutex to avoid concurrent map writes

[eofedorov]

Rutor search was crashing with "fatal error: concurrent map writes" when
the index was updated (loadDB/updateDB) while Search or other readers
accessed shared state (torrs slice and torrsearch index) from other goroutines.

• Introduce a package-level sync.RWMutex and hold it around all reads and
  writes of torrs, isStop, and torrsearch index in Start, Stop, updateDB,
  loadDB, and Search.
• Ensure index is built and assigned under the same lock so no goroutine
  sees a half-updated state.

Add TestConcurrentSearchAndLoadDB to verify there is no data race between
concurrent loadDB and Search; run with: go test -race -count=3.