fix: reset index introduces nan

[sarakolding]

• [x] I have battle-tested on Overtaci (RMAPPS1279)
• [x] At least one of the commits is prefixed with either "fix:" or "feat:"

Notes for reviewers

I think I found out what is happening. Since index is only reset for a single column that is then concatenated back onto the dataset (and I think this is done by matching on indices), a lot of values from the [n_in_bin] column will not have any matches, or will be matched with the wrong bin. I would have done it at the end instead, but the columns that are outputted are also concatenated onto dataframes in the plots, so I would suspect it would only move the issue. Maybe the fix is to reset the index after these steps in the plot functions? You probably have a better overview of what would work best

[sarakolding]

seems to only be a problem for us in plot_metric_by_time_until_diagnosis, not in plot_auc_by_time_from_first_visit, for some reason

[github-actions[bot]]

Tests	Skipped	Failures	Errors	Time
1	0 :zzz:	0 :x:	1 :fire:	5.216s :stopwatch:

[MartinBernstorff]

Huh, that's a bit of a mess! Thanks for taking a look - I'll dive in later today.

[MartinBernstorff]

Since index is only reset for a single column that is then concatenated back onto the dataset (and I think this is done by matching on indices), a lot of values from the [n_in_bin] column will not have any matches

Oh, we need to change this if it happens so that it merges on bin.

I've also added a few tests to hopefully stabilise this interface.