Add single loss evaluation

[MxMstrmn]

Close #90

The single loss evaluation is mostly analogous to the evaluate_r2 function.

• For each combination of drugs, cell line, and dosage we compute the output of the autoencode when presented with the true gene expression values that belong to that combination, y_true.
• The r2_score is computed for each single data point/observation and not only on the mean of the obs that match the drug, dose, cell line ocmbination
• Subsequently, we compute the mean to get one score per combination.

Unfortunately, I overserved many -inf results, so I am excluding them to get proper scores but tracking them during the evaluation.

[MxMstrmn]

Great, thanks @siboehm 🥳 Merging.