[BUG] resolve failing and flaky tests in CI, and race conditions in tests using server

[fkiraly]

Currently, the remote CI is not reliable and will result in faliures and sporadic failures that are unrelated for the PR in question - that means the remote CI is not usable for practical purposes. Problems:

• it may still be that not all failures are temporarily decorated with xfail and skip properly
• I think there is a even race condition in tests for the server, which results in sporadic failures, especially if multiple CI runs happen in parallel (which happens in every PR since there is a test matrix)

I am not sure how to address this best long-term; short-term I would suggest to simply skip all the flaky tests.

Mid-term, I think the server tests need to be decoupled from openml-python SDK tests, i.e., replaced by three things:

• mock backends for simple API tests
• a local server option that runs on the VM. This can follow the extension pattern introduced in the v1-v2 migration, see [ENH] V1 → V2 API Migration #1575
• tests for the server API decoupled from tests from openml-python, possibly on a CRON and controlling for race conditions

[PGijsbers]

Flaky test disabled in #1697: https://github.com/openml/openml-python/actions/runs/22672584548/job/65744252115#step:9:440

FAILED tests/test_runs/test_run.py::TestRun::test_offline_and_online_run_identical - arff.BadLayout: Invalid layout of the ARFF file, at line 0.

[PGijsbers]

tests/test_runs/test_run_functions.py::TestRun::test_local_run_metric_score disabled

FAILED tests/test_runs/test_run_functions.py::TestRun::test_local_run_metric_score - ValueError: Found array with 0 sample(s) (shape=(0, 36)) while a minimum of 1 is required by SimpleImputer.