release: remove test extensions from install-only archives

[indygreg]

These extensions are only needed to support the stdlib tests, which we strip from these archives. These extensions offer no utility. So don't ship them.

This is a follow-up to #1025. Implementing as its own PR since we remove previously existing extension modules. This is backwards incompatible and deserves to be discussed on its own merits.

[geofft]

The breaking change is the unshipping of _ctypes_test, right? That's a very goofy extension that I think has only ever caused our users problems.

[indygreg]

The breaking change is the unshipping of _ctypes_test, right? That's a very goofy extension that I think has only ever caused our users problems.

And unshipping other _test* extensions (like _testbuffer). #1025 is only enabling _testcapi and _testlimitedcapi.

I'm curious how _ctypes_test is causing problems for your users. It was made to build as a shared module in #747. Presumably it is no longer getting imported by anything and not causing the symbol conflicts detailed by #721.

[geofft]

#721 is what I meant - I cannot remember _ctypes_test ever doing anything good, only doing bad things. :) But yeah, I see, the other _test* modules were statically linked.

There's a sort of related question here about whether to ship the Python parts of the test suite (the test module), and I kind of think the answer is we shouldn't in install-only builds. That would also be a breaking change, so maybe there's an argument to do them both in the same release.

FWIW, I'm okay with this personally. On the python-build-standalone side, I don't think there's anything else pending we want to get out the door in the next few days, so we can cut a release with this change alone (maybe plus any PGO stuff you feel is ready). On the uv side, we did just get a breaking uv version bump (0.11.0) out the door today so we're unlikely to do another one any time soon, but we usually don't treat potentially-breaking p-b-s changes as breaking changes in uv, and again this one seems so unlikely to impact real users, so I don't think that should influence anything. But I'll defer to @zanieb and @jjhelmus if they have thoughts.

[zsol]

Out of curiosity, how much do these contribute to the artifact size?