Fix recent issues with CI

[r3kste]

This PR adds fixes for some recent failures in CI.

1. Add pytorch-cpu as a conda dependency, to avoid installation of pytorch-cuda through pip.
2. Ignore mypy warnings for incompatible assignment to pandas DataFrame. An alternative to fix this would be to convert the list to a np.array or pd.Series. However, this might cause unnecessary overhead as internally, pandas doesn't seem to convert the list.
3. Skip a flaky doctest which fails due comparison of positive zero with negative zero, similar to CI failure on 'Run code snippets in documentation' due to negative signed zero. #642.
4. Add a pytest fixture to close matplotlib figures after each test. This is inspired from what is done by seaborn here. This fixture is passed to all functions that create a matploltib figure.
5. Temporarily skip a failing test: tests/optimagic/optimizers/test_pounders_integration.py::test_bntr[start_vec3-cg] on Linux environments with Python 3.10.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
src/optimagic/optimization/history.py	93.63% <100.00%> (ø)
src/optimagic/visualization/slice_plot.py	96.12% <100.00%> (ø)
tests/conftest.py	100.00% <100.00%> (ø)

... and 4 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[timmens]

Thanks for the fix! Can you ignore the failing doctest? We will have to fix this more robustly in the future, but for now that will be fine! Thanks!

[timmens]

Looks great, thank you!!