Add DataFrame fill_nan

[kosiew]

Is your feature request related to a problem or challenge?

There is a common operation in libraries such as PySpark to fill NaN values across an entire DataFrame (or limit by columns). It would be useful to have a similar feature in DataFusion and datafusion-python.

Describe the solution you'd like

If I have a dataframe with a bunch of null values in different columns, I would want to replace all NaNs in those columns with the provided value IF it can be cast to the column's type. Otherwise no-op should happen. Also the user should be able to limit which columns this applies to.

Describe alternatives you've considered

Additional context

This is a repost from apache/datafusion-python#922, prompted by this PR comment

[niebayes]

A possible implementation is to design a bunch of fill UDFs, such as fill_value, fill_prev, fill_linear etc, which act on a column expression in the select list.
The users can then use SELECT fill_value(a, 1.0), fill_prev(b) FROM table to fill selected columns with available fill strategies.
Or use DataFrame::new(..).fill(vec![(Column, FillStrategy), (Column, FillStrategy)]) to apply filling on specified columns of a data frame.