Support Not-NULL Python UDFs

# What
Currently always treat Python UDF return values as "nullable", however that doesn't have to be the case.

# Why
This helps DataFusion to optimize things further.

# How
## WASM UDF
Our WASM UDF interface currently only exposes the [return type](https://docs.rs/datafusion/latest/datafusion/logical_expr/trait.ScalarUDFImpl.html#tymethod.return_type), but not the [return field](https://docs.rs/datafusion/latest/datafusion/logical_expr/trait.ScalarUDFImpl.html#method.return_field_from_args). We would need to wire that up first to support non-nullable return values.

## Python Inference
There are two cases where we can easily infer non-nullable results:

### Python Method w/ Full NULL Handling
Something like:

```python
def add(a: int | None, b: int | None) -> int
if a is None:
a = 0
if b is None:
b = 0
return a + b
```

The user handles the nullability of the arguments and promises to never return NULL.

### Python Method w/ Auto-NULL Handling
Something like:

```python
def add(a: int, b: int) -> int
return a + b
```

The user opted to let us do the NULL handling, i.e. `add` is never called if either of the two inputs in NULL. However if we the two input fields are non-nullable, then we can be sure that the output is also never NULL because we would never skip calling the Python method.

Start with the WASM UDF interface's return_type and return_field_from_args entry points, then trace the Python UDF null-handling and type-inference paths. Done means Python UDFs can expose or infer non-nullable return values, including the full-NULL-handling and auto-NULL-handling cases, and DataFusion receives that nullability.

Written by the indexing model from the issue text.