py-fuzzer: fix minimization logic when --only-new-bugs is passed

[AlexWaygood]

Summary

This PR fixes an issue in the py-fuzzer script described in #17702 (comment). Namely, if run with --only-new-bugs, it's not sufficient for the script to just check that the bug exists in the new version of Ruff/red-knot when the script is minimizing the code that triggers the bug. The script must also check that the minimized version of the code also continues to not cause a panic on the baseline executable if --only-new-bugs is passed. Otherwise, the minimization has gone "too far". It's no longer a new bug if it's been minimized to a snippet that causes a panic on both the test executable and the baseline executable.

The first commit here just simplifies the way we run red-knot on the randomly generated code: the --project CLI argument is no longer necessary, and it's no longer necessary to write a pyproject.toml file to the temporary directory. The second commit here is the one that meaningfully fixes the bug.

Test Plan

I checked that the script continues to pass with mypy using uv run --directory ./python/py-fuzzer --dev mypy from the root of my local Ruff clone.

I also did the following to test this change:

1. I checked out 4a621c2 (the commit prior to 8a6787b)

2. I ran cargo build -p red_knot --target-dir ./target/baseline to create a baseline executable.

3. I checked out this branch again.

4. I ran the following command:

   uvx \                                           
   --python=3.13 \
   --force-reinstall \
   --from=./python/py-fuzzer \
   fuzz \
   --baseline-executable="./target/baseline/debug/red_knot" \
   --bin=red_knot \
   --only-new-bugs \
   50
   

   which produced the following output:

   Running `cargo build --release` since no test executable was specified...
   Sequentially running the fuzzer on 1 randomly generated source-code files...
   Ran fuzzer on seed 50                                         [1/1]
   The following code triggers a red-knot panic:
   
   class name_3[name_2: name_0]:
       pass
   name_3.name_4
   assert name_3[name_0]
   import name_0
   
   New bugs found in the following seeds:
   50
   

   I then confirmed that the script had correctly minimized the snippet to something that caused a panic on this branch but had not caused a panic prior to 8a6787b, unlike the buggy behaviour that the script displays on main.

[github-actions]

ruff-ecosystem results

Linter (stable)

✅ ecosystem check detected no linter changes.

Linter (preview)

✅ ecosystem check detected no linter changes.

Formatter (stable)

✅ ecosystem check detected no format changes.

Formatter (preview)

✅ ecosystem check detected no format changes.

[carljm]

Thank you!