AnyLoc: Towards Universal Visual Place Recognition