Establish whether the DataFusion 55 TPC-H off-heap bump is real, and changelog it if so

### Describe the bug

The DataFusion 55 branch raises the off-heap size in `CometTPCHQuerySuite` from 2g to 4g:

```scala
conf.set(MEMORY_OFFHEAP_SIZE.key, "4g")
```

It is not established whether DataFusion 55 genuinely needs more memory for these queries, or whether this is a consequence of the `peak_mem_used` accounting change in apache/datafusion#22710 that also disabled two `CometAggregateSuite` metric tests.

### Expected behavior

If DataFusion 55 really does need more off-heap for TPC-H, that is a user-visible change and belongs in the changelog: anyone who sized their off-heap against TPC-H would be under-provisioned after upgrading. If it is only an accounting artifact, the bump should be reverted along with the metric fix.

### Additional context

Found while reviewing #5262. Related to the ignored aggregate metric tests.

Start with the DataFusion 55 change in CometTPCHQuerySuite where MEMORY_OFFHEAP_SIZE is raised from 2g to 4g, then review the disabled metric tests in CometAggregateSuite and the peak_mem_used accounting change in apache/datafusion#22710. Compare whether TPC-H genuinely requires the larger limit or whether the bump is an accounting artifact; changelog the user-visible increase, or revert it with the metric fix if not.

Written by the indexing model from the issue text.