feat: Support Enum, @staticmethod, @classmethod mutations

[nicklafleur]

Mix of boy-scouting and new features along with docker improvements for non-linux systems.

Summary

• Refactor Config to a singleton pattern — Extract configuration into a dedicated Config class with get()/reset()/ensure_loaded() API, replacing the global mutmut.config (kept behind a deprecation warning).
• Add type annotations throughout and a safe_setproctitle wrapper for macOS Python 3.14+ fork crashes.
• Fix timeout checker bug — Replace the pid-to-mutant lookup (which used a stale loop variable, causing hangs in parallel runs) with a min-heap based design that tracks (deadline, pid) tuples directly.
• Add enum and static/classmethod mutation support — Introduce an external injection pattern for enum class mutation (avoiding metaclass conflicts), allow @staticmethod/@classmethod through the decorator filter, and add pragma: no mutate class/pragma: no mutate function pragmas.
• Refactor mutation internals — Extract pragma_handling.py, enum_mutation.py, trampoline_templates.py, format_utils.py, and timeout.py into focused modules.
• Improve Docker test image — Copy pyproject.toml/uv.lock before uv sync so source changes don't re-download dependencies.
• /scripts/run_tests.sh now accepts multiple python versions for comprehensive testing workflows.

[Otto-AA]

Iirc, the issue is the importing of setproctitle before forking, not the actual calling (or they changed that in some setproctitle version).

If that's the case, we could try to do the import conditionally after the fork happened and still make use of setproctitle.

I can look up the discussions when I have more time. If you want to search it, there's a PR + linked issue about it.

[Otto-AA]

Hi, thanks for the PR! I'll take a look this week, not sure yet when I'll find the time.

Fix timeout checker bug — Replace the pid-to-mutant lookup (which used a stale loop variable, causing hangs in parallel runs) with a min-heap based design that tracks (deadline, pid) tuples directly.

Can you explain which bug this solves? How did the pid lookup become stale (did someone else modify the dict in the same time?) and how does this cause a hang?

[nicklafleur]

Fix timeout checker bug — Replace the pid-to-mutant lookup (which used a stale loop variable, causing hangs in parallel runs) with a min-heap based design that tracks (deadline, pid) tuples directly.

Can you explain which bug this solves? How did the pid lookup become stale (did someone else modify the dict in the same time?) and how does this cause a hang?

Hey sorry I should have included more details in the PR, here's what I added in the commit (49b4a1e).

The old timeout_checker used the loop variable `mutant_name`
from the outer mutants iteration when looking up estimated_time_of_tests,
instead of looking up the mutant name associated with each PID, causing
some tests to hang and never be killed due to incorrect timeout
calculations.

This was easy to miss because the bug only manifested when multiple mutants
were being tested in parallel and the timing was just right for the wrong
mutant's timeout to be checked against another mutant's PID (in my case,
it only happened specifically when running the e2e tests in vscode
after my last set of changes 🤷)

Instead of trying to patch this specific bug (and probably introducing
one), I decided to go with a simpler design that avoids trying to find
the right pid at all, instead using a min-heap to track (timeout, pid)
tuples.