[druid] fix bug around handling NULLs

[mistercrunch]

fillna would miss out on identifying STRING columns for Druid and
replace None in string columns with a numeric 0. This
mixed type column would confuse
pandas down the line on some operations like df.pivot_table.

superseeds #4236

@xrmx 👀

[mistercrunch]

Note that get_fillna_for_columns assumes that the dataframe's column names are bound to table column names and related types. While this is generally the case, this is not enforced, and the bug could show up again if a string column in a dataframe is not named after a database column.

Viz like the Sankey may use specific columns as a source but rename dataframe columns to source and target for instance. In this case the current logic would fail at identifying the df columns as strings and the mixed type issue may re-appear.

[xrmx]

The space was added on purpose for having NULL entries before other values after sorting

[mistercrunch]

gotcha added it back

[mistercrunch]

A more final solution (outside the scope of this bugfix PR) would be to get better at typing dataframes (currently we have a lot of object types depending on the db engine) and pandas is much slower when operating over object types. We should infer type earlier on in the pipeline and have fillna logic be based on dataframe types instead of column types. (though dataframe types can be inferred from the database types in the first place, though drivers/pandas may fail at that). @betodealmeida and I had a conversation about that last week.