More specific logging in `gather_for_metrics` (#1784)
* Start on testing behavior
* Add test to capture current behavior
* Cleanup test; add length to DummyIterableDataset
* Remove wip test from test_dataloader.py
* Only check on remainder state if we're at the end of a dataloader
* Cleanup
* Fix style
* Move test to test_metrics
* Remove 2 num_process assertion so that we test on single-GPU as well,
why not
* Use `isinstance()` instead of `type()` in test_metrics
Co-authored-by: Zach Mueller <muellerzr@gmail.com>
---------
Co-authored-by: Zach Mueller <muellerzr@gmail.com>
