Strengthen weak refs when sorting in-place

[toddr]

In #16162 , We determined to revert f6107ca in commit d830645 because it was breaking MLEHMANN/AnyEvent-HTTP-2.23. The plan was to put the commit back into 5.30 but I think we're too late in that cycle and the current one to put it back in.

So my question: Do we want to action this in v5.33.1 or do we want to drop this all together? What notifications will need to be made to make this happen?

[xsawyerx]

Honestly, I do consider this a bug and I'm unhappy we need to keep a bug because a single person is being stubborn.

[dur-randir]

But we already have a 'sort' pragma that can change behaviour/implementation. Can it be expanded to provide the desired feature? It could be useful at times.

[xsawyerx]

I do have some old insult-laden email exchanges (sigh) which might have had that as a rejected suggestion, but I'm not sure. I'll need to check.

In any case, I think the objection was to force our hand to not make a change the said module author considered unnecessary (at least that's the word I'm going with). I don't think providing another option to achieve would convince said author. (I could be wrong.)