Fix fillna accepting downcast as dict

[phofl]

• closes BUG: AttributeError when using dict downcast in DataFrame.fillna #40809
• tests added / passed
• Ensure all linting tests pass, see here for how to run them
• whatsnew entry

This fix makes sense in this case I think, but this does not fix the case for non-list like values, which raises a NotImplementedError

The docstring for Series.fillna says, that downcast may be a dict too, but I don't think that this makes sense, so we should maybe raise an error there?

[jbrockmendel]

This fix makes sense in this case I think, but this does not fix the case for non-list like values, which raises a NotImplementedError

by coincidence i've been looking at Block.downcast and found that in the 1D case all test cases have dtypes in [None, "infer"] (and the None case defaults to "infer"), and as you noted the 2D case only accepts "infer". Anything we can do to be stricter and/or get downcast out of Block would be a plus by my lights

[phofl]

Could deprecate the downcast keyword?

[jbrockmendel]

Could deprecate the downcast keyword?

id be on board for this, no idea about others.

[jreback]

thanks @phofl

[jreback]

Could deprecate the downcast keyword?

id be on board for this, no idea about others.

yeah +1 here as well. this was an artifact from a while back when this was expensive. we should always just downcast (as that is what we do everywhere else)

[phofl]

Open an issue about this or directly a pr?

[jbrockmendel]

Open an issue about this or directly a pr?

probably merits an issue, since im not sure @jreback and i are on the same page. my immediate interest is in getting the downcast keyword out of Block.fillna (and ideally Block.interpolate), which i think is distinct from what he has in mind

[phofl]

opened #40988