gh-144278: Enable overriding TAG (sys.implementation.cache_tag) when building sysmodule.c

[zooba]

In particular, this allows defining in to NULL, which has the effect of disabling automatic .pyc caching.
We also update tests to handle the cache tag being None, which was always allowed, but was not accounted for in tests.

• Issue: Allow disabling cache_tag at build #144278

[zooba]

If anyone stumbles into here and has any idea why execution is 2-3x slower when cache_tag is None, I'd love to hear about it. The test failures/cancellations are timeouts/time-sensitivity issues, not poor functionality, but I can't come up with any reason why it would be so much slower.

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @zooba for commit c277edb 🤖

Results will be shown at:

https://buildbot.python.org/all/#/grid?branch=refs%2Fpull%2F144293%2Fmerge

If you want to schedule another build, you need to add the 🔨 test-with-buildbots label again.

[zooba]

So it's looking like our compiler is now slow enough that the test suite can't finish within timeouts without cached bytecode. Running with the buildbots for a broader view including PGO/etc., but it's possible the answer is just going to have to be to run with longer timeouts to see the results. Or optimise the compiler a bit.

FWIW, on my own (fairly slow) Windows PC, the difference isn't anywhere near as bad as in CI.