[primer] Flaky results from inference differences

[jacobtylerwalls]

See #8758 (comment) for a recent example of subtle differences in messages comparing unequal and causing noisy, irrelevant primer output.

Messages like invalid-name and deprecated-typing-alias seem to be the worst offenders.

[jacobtylerwalls]

I'll reopen if I see this again.

[Pierre-Sassoulas]

I had a long term plan to make the primer prettier by showing only what really changed, but my work branch is basically impossible to rebase at this point. Should have concluded the MR when I had the chance. It's not super simple to do, you need to decide if a message is a new message or a modified message from information that is not very stable. Basically some fields are more important than others. Also the commit is not set for the primed repo, which make this a shaky ground to decide from the json we have. And there's of course a lot of markdown output fine tunning to do.

[jacobtylerwalls]

Still happening, not particularly subtle.

See recent runs like #8824 (comment), on a PR only touching pyreverse 😭 .

[jacobtylerwalls]

I have reason to believe #8872 made this significantly more deterministic. Will reopen if I see it again.