Enhance udf to take additional non-expr arguments

[timsaucer]

Is your feature request related to a problem or challenge? Please describe what you are trying to do.

There are some cases where you would like to pass additional data to a UDF that are not expressions. For example, suppose you wrote a UDF that did some kind of lookup and the function had a signature like

def my_lookup(column: Expr, lookup_values: dict[str, str]) -> Expr

It would be convenient to allow for passing in arbitrary data.

Describe the solution you'd like

Most likely this would mean a change to PyScalarUDF::__call__ and I could see changing the function definition to something like fn __call__(&self, args: Vec<PyAny>) -> PyResult<PyExpr>. It's not immediately obvious to me how to set additional parameters, but it may mean switching from SimpleScalarUDF to a different ScalarUDFImpl to carry the additional data.

Describe alternatives you've considered

Right now you would need to do one of these approaches

• create a class that you can set these values on and then use the __call__ function as your UDF
• take advantage of setting the variable scope for lookup_values
• Pass the variables as literals

Additional context

[timsaucer]

This was implemented in #880 for window functions, just need to add it in for UDAF/UDF

[timsaucer]

After some initial investigation, the way the window functions are using an instance has a problem in that when the instance is reused it's carrying over the past state. I bet that's why the aggregate functions were set up to use a class type and then instantiate them in to_rust_accumulator. I'm trying to understand why this does work because it seems like the only place we instantiate is during the create_udaf however we're clearly getting two calls to __init__ in the unit test when I add a second aggregation in the same test using the same udf. This requires further investigation.