Dataframe doesn't properly implement ArrowStream export interface

[johnnyg]

Describe the bug
When trying to pass a dataframe to another library that expects an ArrowStream export interface we get the following error:

TypeError: argument 'input': DataFrame.__arrow_c_stream__() missing 1 required positional argument: 'requested_schema'

this is because the requested_schema argument should be optional but it's not.

To Reproduce

from arro3.core import RecordBatchReader
import datafusion

data = [{"num": 42}]
ctx = datafusion.SessionContext()
df = ctx.from_pylist(data)
reader = RecordBatchReader.from_arrow(df)

Expected behavior
The above to run without error

Additional context
Replacing RecordBatchReader.from_arrow(df) with RecordBatchReader.from_arrow(df.df) works around the bug

[kylebarron]

Additional context Replacing RecordBatchReader.from_arrow(df) with RecordBatchReader.from_arrow(df.df) works around the bug

Ah so it's not the Rust side, which implements it correctly:

datafusion-python/src/dataframe.rs

Lines 841 to 846 in 98dc06b

	#[pyo3(signature = (requested_schema=None))]
	fn __arrow_c_stream__<'py>(
	&'py mut self,
	py: Python<'py>,
	requested_schema: Option<Bound<'py, PyCapsule>>,
	) -> PyDataFusionResult<Bound<'py, PyCapsule>> {

It's on the Python side:

datafusion-python/python/datafusion/dataframe.py

Line 1056 in 98dc06b

def __arrow_c_stream__(self, requested_schema: pa.Schema) -> Any: