Should we continue to disallow nans in predictions?

[phinate]

https://github.com/alan-turing-institute/cloudcasting/blob/d84dde0a2f115dc6d7f9f6dcbd5ff0d951b210b9/src/cloudcasting/models.py#L41-L43

Not fully convinced we need this now that our metrics handle NaNs gracefully.