Requirement for nullable test data

[glatterf42]

Not sure this is an actual issue, this needs some clarification first.

As part of #41, I tried updating the pandas version to 2.2.0. This showed numerous DeprecationWarnings and FutureWarnings, one of which was related to .fillna(np.nan) and can be found in pandas release notes. My solution was to simply remove the fillna(np.nan) part of the offending line since this did not affect our results. As @meksor pointed out, we should however check if this is a requirement we want to keep and clarify what exactly this requirement is. Does this only work because our test data only contain np.nan whenever a data cell in a table would be empty? Is there some implicit conversion going on in some background step? For example, pandas.read_sql() has a parameter called dtype_backend, which defaults to "numpy_nullable". Does this convert missing values to np.nan?

Possibly related: a migration guide to copy-on-write behavior.