Convert neuralop model checkpoints for version >0.2.0

[arthurfeeney]

It seems that neuralop refactored some of their internal data structures, so checkpoints that were trained with version 0.2.0 are not loading when using neuralop's main branch. This may only be the case when using the tuckerized version... :shrug: Not terrible, can just set 0.2.0 as the version w/ conda, but would be nice to work with later versions.

There may be a way to convert the checkpoints to the new version. Probably easiest to ask them in an issue instead.