Fix context bug with CONDA_REPODATA_FNS and improve some tests in test_solve

[jaimergp]

Description

Some tests in test_solve were using the cached solver backend, so the solver parametrization was not working properly.

Additionally, we explicitly set CONDA_REPODATA_FNS so libmamba can recognize it was asked for (and not injected in conda.cli.install).

While doing this I discovered that the current conda will crash if CONDA_REPODATA_FNS is set to something not including repodata.json because a list is expected but context returns a tuple.

Checklist - did you ...

• Add a file to the news directory (using the template) for the next release's release notes?
• Add / update necessary tests?
• Add / update outdated documentation?

[codspeed-hq]

CodSpeed Performance Report

Merging #13490 will create unknown performance changes

_{Comparing jaimergp:test-solve-fixes (543ec7d) with main (7ed9d45)}

Summary

⚠️ No benchmarks were detected in both the base of the PR and the PR.

[jaimergp]

4084.33s call     tests/test_create.py::test_install_features[classic]

o.o WHAT

[jezdez]

@jaimergp Any idea where the 3.12 test failures comes from?

[jaimergp]

I'm looking into it. I can reproduce locally and only with 3.12. Very intriguing. Also the test is slow. 4000s in CI, 1000s locally o.o

[jaimergp]

I think it's an interaction between main and conda-forge that makes the classic solver faint :P These integration tests shouldn't assume compatibility with the pre-configured channels when issuing a --channel=xxx argument, so I added an --override-channels flag and now it passes (and super fast!).

[jaimergp]

Passing now @jezdez