Re-think use of MultiIndex in `to_xarray_dataset`

[samantha-ho]

The to_xarray_dataset method, by default, will create a MultiIndex if there are repeated values along any axis, and will then convert all coords to data_vars. This can cause problems in the expected xarray dataset structure.

We should re-think what cases really need a pandas index or a MultiIndex to minimize these unexpected issues.

[jenshnielsen]

I think part of the problem is that a interrupted dataset returns only the created data from get_parameter_data this means that trying to reshape the data into the correct shape will always fail and the fallback will use MultiIndex. It might make more sense if the data was padded with NaN or similar for the missing data.