Consolidate numpy conversion functions

[dfulu]

Currently we have a whole suite of numpy conversion functions in the numpy_sample module. In the samplers we convert each modality into a dictionary separately and then merge the dictionaries.

I think it could be cleaner to create a single function which replaces all the numpy conversion functions in the numpy_sample module. As part of this I would also suggest that we should remove all of the {X}SampleKey classes scattered throughout the numpy_sample module. Instead all of the keys of the returned dictionary would be obvious from the new single numpy conversion function.

The new conversion function could accept a dictionary of xarray objects identical to the dictionary we pass around inside the PVNet Dataset classes to slice time and space components. We'd need to decide a little more on how to add t0, the time encodings and the solar coords to the sample, but I feel like this should be reasonably straight forward

[yuvraajnarula]

@dfulu if this issue is not resolved, I would like to contribute.

[dfulu]

Hi @yuvraajnarula thanks for the offer. This issue is blocked now until after #375 is finished and merged