TabularData: Use `None` internally

[daavoo]

After reading your explanation, I think this PR is probably fine for now, maybe just open a separate issue regarding making TabularData store None types instead of the fill strings.

Originally posted by @pmrowla in #7064 (comment)

Use None internally and only replace it with fill_value when rendering.

[karajan1001]

In pandas Dataframe, the behavior is we don't give a default value to replace null when the Dataframe is created.
But the calling of method fillna will permanently replace the null value.

[skshetry]

Related: #6893 (comment) and #6867 (comment).

It's not just None. Ideally, I'd like to have support for arbitrary data. This will come really useful for example in floats rendering.
We may need to do some filtering based on actual floats' values and/or need to have different forms of rendering them ("%g" vs "%e" etc.) and with different precisions.

The internals has to change significantly I think to make these adjustments, and may probably be the time for it.