[python] TileDB-SOMA-Py can do better null-filtering with implicit Pandas DataFrame categories

[johnkerl]

Split out from #2858.

Here is a repro script: https://gist.github.com/johnkerl/68c978ac9d3e774749f77e704fd718d3

Here is a readback script: https://gist.github.com/johnkerl/20e0ad08701f5913f90be706ecd99b01

Notes;

• In either case, the input values are ["", "B cell", "T cell", None, pd.NA, math.nan]
• If the pd.Categorical levels are supplied as ["B cell", "T cell"] then all is well
• But if the pd.Categorical levels are not supplied, we get six levels:"B cell","T cell","","None","", and"nan"`
• In this latter case:
  • We should proactively drop None, pd.NA, and math.nan from any provided levels. It makes no sense to do anything else -- ?
  • On the other hand, compbios may wish to preserve a distinction between None (missing) and NA (not missing, and known to be not applicable)