Caching for OT loss metrics

[ehuan2]

Currently, as shown in: https://github.com/ehuan2/crispy-fishstick/blob/281a36ed2065c41b26ceb99f6edab05d4b45d6ec/src/crispy_fishstick/metrics/gex_prediction/ot_eval/base.py#L46 we don't cache the dataset to memory which could add on significant time if we're not dealing with a small dataset.

We can likely change this to a separate cache with this comment here: #37 (comment)

[aosakwe]

ah I had missed adding that - I can assign this to myself and do it