perf: coalesce batches before sending to distributor channels in RepartitionExec (#22010)

## Which issue does this PR close?

<!--
We generally require a GitHub issue to be filed for all bug fixes and
enhancements and this helps us generate change logs for our releases.
You can link an issue to this PR using the GitHub syntax. For example
`Closes #123` indicates that this PR will close issue #123.
-->

- Closes #22202

## Rationale for this change

<!--
Why are you proposing this change? If this is already explained clearly
in the issue then this section is not needed.
Explaining clearly why changes are proposed helps reviewers understand
your changes and offer better suggestions for fixes.
-->

Today RepartitionExec emits one small batch per (input batch × non-empty
output partition), then coalesces those small batches back into
target-sized ones on the consumer side. That means the channel layer
(memory accounting, gate, await suspensions) does work proportional to
num_partitions per input batch, even though each small batch only
carries batch_size / num_partitions rows. Moving the coalescing
producer-side, before the gate, collapses the channel traffic and
reduces sensitivity to high output fanout.

This gets much worst in
https://github.com/datafusion-contrib/datafusion-distributed, which uses
`RepartitionExec` as the foundational backbone for network shuffles: it
scales up the `RepartitionExec` output partitions to match P
(partitions) * W (workers), where P is typically in the range of 12-24,
and W can be from 1 to thousands. The fanout overhead that the current
`RepartitionExec` introduces is pretty big in that setup.

Ideally, the overhead of fanning many small `RecordBatches` should be as
small as possible.

## What changes are included in this PR?

<!--
There is no need to duplicate the description in the issue here but it
is sometimes worth providing a summary of the individual changes in this
PR.
-->

- A producer-side coalescer per output partition, shared across input
tasks, that runs before the channel's gate.
- The redundant consumer-side coalescer in PerPartitionStream is
removed.
- BatchPartitioner is unchanged — only the layer between the partitioner
and the channel.

## Are these changes tested?

<!--
We typically require tests for all PRs in order to:
1. Prevent the code from being accidentally broken by subsequent changes
2. Serve as another way to document the expected behavior of the code

If tests are not included in your PR, please explain why (for example,
are they covered by existing tests)?
-->

Yes, by existing tests

## Are there any user-facing changes?

No

<!--
If there are user-facing changes then we may require documentation to be
updated before approving the PR.
-->

<!--
If there are any breaking changes to public APIs, please add the `api
change` label.
-->

Author: gabotechs