Magic v2 Internal Code Conversions

[leonerd]

(This is a dependent PR that requires first that #24163 be merged)

The three four new commits here are each individual conversions of static internal magic types to use magic v2. None of them are strictly "required", but we felt it useful to have some code sitting around that uses magic-v2, as a demonstration that it works. These ones were picked because they're all static and don't have associated PERL_MAGIC_* constants anyway, so there's no way that external code / CPAN / DarkPAN would be aware of them in any way that this could break.

[leonerd]

All three commits currently introduce new warnings when building with g++, as discussed at some length over in #24508 (comment) and which ultimately lead to the mailing list discussion about using g++ in the first place:
https://www.nntp.perl.org/group/perl.perl5.porters/2026/06/msg271009.html

I still don't have a satisfactory solution to this problem. My personal vote is "lets stop caring about building Perl using g++".

[tonycoz]

Another one would be argvout_vtbl in doio.c, that's on an IO but it only uses the free and dup callbacks.

[tonycoz]

It looks reasonable, but I think the g++ build thing needs to be resolved one way or another.

[leonerd]

I've opened a new issue (#24553) to discuss the g++ warnings issue

[leonerd]

Ahyes; I had missed the one in doio.c before, but now added. I don't know why it had previously been using PERL_MAGIC_uvar as it didn't seem to be using any of the special get/set behaviour that provides, just free and dup. So it might as well be a MGv2s_BASE now.

[Leont]

:encoding has always struck me as the sort of module that should be dual-life, but clearly absolutely no one wants that responsibility so probably this is fine.

[leonerd]

Annoyingly, github appears to believe this PR wants to add 6 commits, even though the first 2 of those have already been merged; this would only do the other 4. Rather than risk an upset here I will merge it on commandline.

[leonerd]

github is confused about what I did here. This was merged with a local git merge --ff-only, to preserve the identity of each of the 4 commits