fix: Correctly handle signature-less functions for Py UDF calls

[jmao-denver]

Fixes #6349

[chipkent]

This PR is marked as fixing #6349. #6349 mentions two problems. This test is casting the return type, so it is either a partial fix of the ticket or there should be another test to check type inference.

[chipkent]

Looking into this problem some more, pybind11 actually does provide signature information in a way that I think we are not yet supporting. From the example in #6349, I added:

print(blackscholes.price.__doc__)

This outputs a docstring that is prefixed with the method signature.

price(arg0: float, arg1: float, arg2: float, arg3: float, arg4: float, arg5: bool, arg6: bool) -> float

Here are more things that may be useful:

print(type(blackscholes.price))

<class 'builtin_function_or_method'>

print(blackscholes.price.__qualname__)

PyCapsule.price

[jmao-denver]

1. I don't think the doc string of a callable is a reliable source to infer its signature. Even if it is, the work needed to do that correctly may not justify the benefit
2. that said, jedi does try to infer function arguments for sphinx, epydoc and basic numpydoc docstrings, but it doesn't expose any public API for that.
3. PyCharm seems to rely solely on signatures to do static check/code assistant on functions

Based on the above, I am not sure it is so bad that we just document the limitation and workaround when we can't get a signature via the standard inspect module.

[chipkent]

We could easily make this "just work" by using the return type. Relying on documented workarounds is a poor substitute to the product functioning as expected.