[Python] Allow constructing InMemoryDataset from RecordBatchReader

[amoeba]

Describe the bug, including details regarding any error messages, version, and platform.

The docstring for InMemoryDataset indicate you can create one from a RecordBatchReader:

arrow/python/pyarrow/_dataset.pyx

Lines 990 to 999 in 5a52916

	cdef class InMemoryDataset(Dataset):
	"""
	A Dataset wrapping in-memory data.
	Parameters
	----------
	source : RecordBatch, Table, list, tuple
	The data for this dataset. Can be a RecordBatch, Table, list of
	RecordBatch/Table, iterable of RecordBatch, or a RecordBatchReader
	If an iterable is provided, the schema must also be provided.

However, if you try this you currently get an error saying you cannot:

>>> ds = ds.InMemoryDataset(rbr)
Traceback (most recent call last):
  File "<python-input-37>", line 1, in <module>
    ds = ds.InMemoryDataset(rbr)
  File "pyarrow/_dataset.pyx", line 1038, in pyarrow._dataset.InMemoryDataset.__init__
TypeError: Expected a table, batch, or list of tables/batches instead of the given type: RecordBatchReader

I don't think we allow simple construction of an InMemoryDataset from a RecordBatchReader because that violates the assumption Datasets about sources being re-readable (not one-shot like RBR). But I don't see why the InMemoryDataset constructor can't consume the RecordBatchReader and construct a Table from it.

Component(s)

Python

[amoeba]

Issue resolved by pull request 46731
#46731