[C++][Dataset] Separate datasets backed by readers from InMemoryDataset

[asfimport]

From ARROW-10882/#9802 

• Backing an InMemoryDataset with a reader is misleading. Let's split that out into a separate class.
• Dataset scanning can then use an I/O thread for the new class. (Note that for Python, we'll need to be careful to release the GIL before any operations so that the I/O thread can acquire the GIL to call into the underlying Python reader/file object.)
• Longer-term, we should interface with Python's async.

Reporter: Weston Pace / @westonpace
Assignee: David Li / @lidavidm

Related issues:

• [Python][Dataset] Writing dataset from python iterator of record batches (relates to)

PRs and other links:

• GitHub Pull Request #10070

_{Note: This issue was originally created as ARROW-12231. Please see the migration documentation for further details.}

[asfimport]

Weston Pace / @westonpace:
To add to this.  The reader-backed dataset is not restartable while a RecordBatchVector/Table-backed dataset could be restartable, or it could remove its ownership of the record batches / table as it scans (allowing batches already scanned to be freed).

[asfimport]

Ben Kietzman / @bkietz:
Issue resolved by pull request 10070
#10070