Should `open_mfdataset` try to guess which engine to use?

[jsignell]

What is your issue?

The guess_can_open logic in backends can be hard to get right. It is certainly helpful to be able to guess an engine when opening a single file, but when you are opening a whole bunch of files you probably should explicitly set the engine.

I guess the action on this ticket would be to make the engine kwarg required on open_mfdataset.

[shoyer]

I think it's OK to guess which engine to use, but it should probably only happen by looking at the first file, not all of them.