Check if evaluation fails if examples or ground_truth have all `None`values

[javiermtorres]

If examples or ground_truth values are all None, evaluations could fail due to an incorrectly inferred format in HF datasets. This needs to be checked at some point.

Context:

Maybe None in examples or ground_truth would need to be handled differently - like should those instances be dropped from the dataset itself? WDYT?
Its a different issue than what you are addressing here, so we can tackle it else where :)

Originally posted by @HareeshBahuleyan in #1265 (comment)