API break with DataFrame.rename

[TomAugspurger]

Posted in #17800 (comment), but making a new issue to make sure it's not lost and that everone is aware, and to check if we're comfortable with this:

df = pd.DataFrame(columns=['A', 'B'])
df.rename(None, str.lower)

In previous versions that meant .rename(index=None, columns=str.lower), but now it's .rename(mapper=None, index=str.lower). I think this is fundamentally ambiguous correct? This is breaking dask, which used positional arguments, but I suspect it'll break more code.

cc @jreback @jorisvandenbossche

[TomAugspurger]

Thinking out loud here:

We could change the signature of DataFrame.rename to (*args, **kwargs). Then

1. Call with all keyword arguments, no ambiguity (ideal case)
2. Call with positional arguments. We interpret that as .rename(index=args[0], columns=args[1]) and warn telling them to use keyword args
3. Call with 1 positional argument, and an axis= kwarg. No problem
4. Everything else raises?

downside: we'll need to rewrite DataFrame.__signature__ so that introspection works properly, won't be too bad. Let me see if that works.