Learning And Control Of Deep Koopman Eigenfunctions