String dtype: allow string dtype in query/eval with default numexpr engine

[jorisvandenbossche]

Currently (on main, with no future options enabled), you can use query on an object-dtype column of strings. For exeample:

>>> df = pd.DataFrame({"col" : ['a', 'b', 'c']})
>>> df.query("col < 'b'") 
  col
0   a

This works despite that the default engine is "numexpr" (when installed) and that this engine does not actually evaluate the above expression. We do have checks to essentially fall back to evaluate the expression in python if one of the operands is object dtype (and the operator is one of the allowed operators for that).

But with the future default string dtype, we trigger the warning about explicitly falling back to the python engine:

>>> pd.options.future.infer_string = True
>>> df = pd.DataFrame({"col" : ['a', 'b', 'c']})
>>> df.query("col < 'b'")
RuntimeWarning: Engine has switched to 'python' because numexpr does not support extension array dtypes.
Please set your engine to python manually.
  col
0   a

Given that we were already automatically falling back for object dtype strings, I would propose to keep doing that for the future default string dtype as well.
It is a bit misleading (you might think to get a speedup when using this, which won't be the case), but that is already the case right now anyway, and otherwise a lot of users will start to see that warning (and a lot of people also use query for the convenience of writing the expression, I think, and not only for the speed).

xref #54792

[mroeschke]

Thanks @jorisvandenbossche