Add dataset class to train LodeRunner on one of the MPP datasets

[schakrabarti-lanl]

This is a dataclass that will be used to train LodeRunner on one of the PDEBench datasets.

[hickmank]

Nice. A few comments, marking "request changes" since it's a draft. I guess the next check would be to see if LodeRunner can actually be evaluated on an output of this dataset class.

[hickmank]

If this is copied from the MPP repo we should not include it in Yoke. If we need the functionality we could add their module as an optional dependency in the pyproject.toml.

[hickmank]

This looks like a great start. We'll have to think about how the in_vars and out_vars inputs to LodeRunner should be handled when training off of this data. But this looks close to something usable!

[hickmank]

@schakrabarti-lanl Awesome Sharmistha, thanks for working on this. It doesn't look too far away from being able to do a test training run of the LodeRunner architecture on this dataset!