issues with `test_solve_regression`

[bouweandela]

The test packages/climate-ref-esmvaltool/tests/unit/test_solve_regression.py::test_solve_regression seems to pick up the default default_ignore_datasets.yml file instead of the one from the git repository it is run from. For example, the default ignore datasets file was updated in #585 without updating the regression outputs, and the tests were all green on that pull request. On a subsequent pull request they fail: https://github.com/Climate-REF/climate-ref/actions/runs/22848594747/job/66271138543?pr=589.

The test is also parametrized in a way that is somewhat confusing and may be needlessly slow. It solves all diagnostics at once and is parametrized only over comparing the results of the solve per diagnostic, instead of solving just the for the requested diagnostic. For example, when running pytest packages/climate-ref-esmvaltool/tests/unit/test_solve_regression.py::test_solve_regression[sea-ice-sensitivity], it will not just solve the sea-ice-sensitivity diagnostic, but also all of the other ones. That is confusing if the solve fails on a diagnostic that is not sea-ice-sensitivity.

[lewisjared]

The test is also parametrized in a way that is somewhat confusing and may be needlessly slow

Good point. We can rework the fixture. I believe I implemented it that way as an optimisation as the results were needed in other places, but can take a look at how we can decompose it in a way to not massively increasing the whole suites run-time.

[lewisjared]

The ignore_datasets caching might need some rework as there is a side-effect on loading the config to access the internet. Also this side effect causes some issues when running in docker if the users home dir isn't writable (default).

Perhaps we need to remove the side effect and move it into a explicit step in the solve process?
If a user specifies a file location they won't get the benefit of fetching.
The config value is where the file should be put and should be decoupled from the fetching lifecycle.