Late file sanity checks on file copy (Rcp, Scp) ?

As of 1.6, ClusterShell performs some basic file sanity check (like : is a directory) on EngineClient initialization. We should probably move these checks to `Rcp._start()` so that they will be performed later, but the exception should be well handled and the user notified as needed. Not sure this is better - this ticket has been created in order to track this possible change.
