Revert PR #9818 (GroupedIterator improvements)

[SethTisue]

This reverts commit d047407.

Because it caused community build failures; see scala/community-build#1519 for details

[som-snytt]

I commented on the PR that the community build just uncovers a bad test or assumption about the API. Make that two bad tests. I only looked at twitter-util which was straightforward.

[SethTisue]

Regardless, I'm still much more comfortable reverting the PR for 2.13.8, in keeping with the informal alternation we tend to have between big releases with months of changes (e.g. 2.13.7) and modest releases focused on addressing regressions (e.g. 2.13.8).

[lrytz]

... I agree.

[joroKr21]

I agree with the revert. It's reasonable to expect that smallCollection.grouped(hugeNumber) won't actually allocate much memory. It's an edge case but a valid edge case IMO (same as collection.take(negativeNumber)). I'm now thinking about another approach where we reuse the initial buffer for everything. That should both reduce allocations overall and allow us to let it grow since it would be a one-time cost. I don't see a theoretical reason why we can't do that.