duplicate scores in database

this is worked around with https://github.com/brain-score/brain-score.web/pull/37 but we seem to have duplicate scores in the database `(benchmark_identifier, model)`. They are completely identical, e.g. for `mobilenet_v2_1.0_224`.

We should:
1. delete the duplicates from the database
2. prevent future duplicates from being added. I think a unique constraint `(benchmark_identifier, model)` might not work because we are adding `null` scores for erroneous runs and need a way of providing an updated score for them. We might already be doing this (https://github.com/brain-score/brain-score/blob/fcea6e19425fe256e69a5ba60580fe0e54225368/brainscore/submission/evaluation.py#L113), but we should verify
3. make sure that when there is a `null` score from an erroneous run and a newer actual score, we use the newer score for display and parent aggregation

Start by reviewing the workaround in brain-score.web pull request 37 and the referenced brainscore/submission/evaluation.py section around line 113. Trace how duplicate database records, null scores, display selection, and parent aggregation are handled; done means duplicates are cleaned up, future additions are prevented, and newer actual scores take precedence over erroneous null scores.

Written by the indexing model from the issue text.