Add pinned dev version of ast_serialize to CI

[ilevkivskyi]

I would propose to merge new-parser in master soon, so that we can parallelize fixing remaining failing tests. This is a preliminary step to check if we can use pinned ast_seriaize in CI.

cc @JukkaL

[JukkaL]

I started preparing the new parser branch for merging, I hope to create a PR later this week.

[ilevkivskyi]

@JukkaL Do you need this CI setup? It looks it is finally working now. This now installs a pinned version of ast_serialize on all Ubuntu workers. I guess you can use this to actually run a selection of tests using the new parser.

[JukkaL]

Without this the new parser tests would have to be run manually, so this could be quite helpful, but I need to think more about this.

[ilevkivskyi]

@JukkaL

Without this the new parser tests would have to be run manually, so this could be quite helpful, but I need to think more about this.

OK, you can take this as building block and adjust as needed. For example run a dedicated job for new parser. Btw for now I am skipping free-threaded build as it requires extra configuration (we can still build the wheel when setting up proper CI in ast_serialize).

[ilevkivskyi]

@JukkaL Now this PR makes parallel tests run with native parser (I temporarily skip parallel tests on Windows). This will:

• Prevent accidental regressions in native parser
• Allow me to start working on making parallel checking native-parser-only.

In parallel we can gradually re-enable remaining ~50 skipped tests. (FWIW I don't think there is anything essential there, mostly different error messages, and some rare edge cases.)

If you don't have objections, I will merge this tomorrow.

[github-actions]

According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅