Testing improvements

[maxrjones]

What is your issue?

As mentioned in #71 (review), the xbatcher test suite needs improvements. This issue will track necessary improvements to the tests (additions are welcome):

• Test data loaders on both datasets and dataarrays (done for pytorch at Let torch accessor and dataloader handle either xarray.DataArray or xarray.Dataset inputs #85, still need to do for keras/tensorflow)
• Make data loader tests more granular
• Test batch generation on datasets with multi-dimensional coordinates
• Test batch generation on datasets with non-dimensional coordinates
• Test batch generation on datasets with other index types
• Use xarray testing methods on batches for more informative diffs
• Add a test for batch_dims (Batch generation with batch_dims in v0.2.0 is about 10-20times slower #121)