DiskDataset assumes the X field is populated, in several places

While it is possible to easily create a DiskDataset that does not have the X field populated, there are lots of places in the DiskDataset code that assume that it is.

There is a strong use case for being able to make DiskDataset objects with, say, just a y field. For doing inference on large datasets, it makes sense to create a DiskDataset to contain the resulting generated data, and the most reasonable place to put this is in the y field.

There are several places in the code this assumption is made; for example in get_shard() it's assumed that the X file is there. Notably, this is not checked for on DiskDataset creation.

Begin at DiskDataset creation and get_shard(), then locate the other DiskDataset paths that assume an X file exists. Done means a DiskDataset containing only y can be created and used through those paths without requiring X, with regression coverage for the behavior.

Written by the indexing model from the issue text.