extractor: add explicit numerical columns tconvert_columns_to_category

[awillecke]

The PlottingReaderFeather converts certain columns to categories automatically. numerical_columns should be used to specify which columns should not be converted, when reading data. If, however, a column is already categorial, it will not be converted to numerical explicitly, which does not function as expected. This commit adds the explicit conversion to float for numerical_columns .

[hagau]

Nice, thank you! numerical_columns is indeed more expressive.
I'll merge this as is, since it is an improvement over the existing code, but I think I'll need to change the convert_columns_to_category in the near future.
In most cases the conversion to float will be correct, but in some cases other data types will be called for. This can be achieved with a dictionary of pairs of column names and data types to override the implicit type casting done by pandas. This can be optional and the conversion to float can be the default if a set is given instead of a dictionary.
Instead of excluded_columns it's probably better to explicitly name the categorical columns, since constructing the set for the heuristic used right now is somewhat costly with regard to memory and time for larger DataFrames.

[awillecke]

Yeah, that is a good idea to provide a data type dictionary explicitly when extracting data 👍 I created a draft that I'm currently working on. Maybe it is a good place for integrating and testing.

[hagau]

@awillecke Yes, good idea. I'll use that as a basis.