test: replace viser-based setuptools override repro with minimal fixture

[baszalmstra]

Description

test_setuptools_override_failure has been using viser==0.2.7 as its reproducer since #1969, purely because that package's 52-distribution transitive graph happened to trigger the bug from #1686 (the resolver's in-memory index leaking into build-dispatch). That graph is expensive: ~42-50 s wall on Windows, and pulls cmake, ninja, xatlas, and assorted unrelated dependencies solely as side effects.

This PR replaces the fixture with a trivial local path-based pypi package that uses setuptools.build_meta as its build backend. No production code changes. The test keeps the same name and still fails if the #1969 fix is reverted.

Why the replacement still reproduces #1686:

• setuptools is listed as a conda dependency, so the resolver emits its "assumed to be installed by conda" override for it (log confirms: the following python packages are assumed to be installed by conda: wheel ..., pip ..., packaging ..., setuptools ...).
• The local package's PEP 517 build triggers build-dispatch, which must install setuptools>=60 from PyPI.
• Before fix: use two in memory indexes, for resolve and builds #1969, build-dispatch shared the resolver's in-memory index and would have seen the conda override instead of a real PyPI wheel. Current code forks shared_state so build-dispatch has its own index and succeeds.

Measured on Windows, warm cache, PIXI_CACHE_DIR=<fresh tmpdir> (same isolation pattern as before):

Fixture	Wall time
viser (before)	42-50 s
minimal (after)	~17 s

How Has This Been Tested?

Ran the replaced test locally on Windows with the slow_integration_tests + online_tests features, on a fresh PIXI_CACHE_DIR tempdir:

cargo nextest run -p pixi --test integration_rust \
  --features slow_integration_tests,online_tests \
  -E 'test(install_tests::test_setuptools_override_failure)'

Passes in ~17 s. Logs confirm the setuptools override is emitted by the resolver and that the PEP 517 build runs through build_dispatch without failure.

AI Disclosure

• This PR contains AI-generated content.
  • I have tested any AI-generated content in my PR.
  • I take responsibility for any AI-generated content in my PR.

Tools: Claude Code (Opus 4.7)

Checklist:

• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• I have added sufficient tests to cover my changes.