⚖️📄 Compare with reproduction results, if available

[mberr]

This PR compares the results obtained by the reproducibility pipeline to those present in the configuration

• extract metric names & values from configuration
• PipelineResult.get_metric cannot resolve some metric names as present in the configuration, e.g., "hits_at_k.best.10". Update metric name resolution
• create and store a dataframe with the original and reproduced metric values
• summarize difference with significance check

[mberr]

@cthoyt any thoughts about the metric name resolution?

[mberr]

I made a new utility in b6132f3 (functional, i.e. we do not need to instantiate a RankBasedMetricResults)

[cthoyt]

I think I'd go for updating to be able to handle hits_at_k.best.10... not sure how those got out of sync

[cthoyt]

can you show what the output looks like now when this is run?

[cthoyt]

needs an emoji, then merge when ready

[cthoyt]

LSTM!