Stabilize the ignored-keyword warning test across CI

[cvanelteren]

ultraplot/tests/test_format.py contains a skipped test_ignored_keywords() with the note that it fails on GitHub but not locally. The warning paths themselves are real, but the test currently looks brittle.

This issue is intentionally scoped for newcomers to the project. It is low complexity, touches a single test, and is a good way to get familiar with UltraPlot's warning behavior without changing core plotting logic.

Suggested scope:

• inspect why the current test is environment-sensitive
• replace raw warning-count assertions with more specific checks on the relevant warning messages
• keep the test focused on gridspec_kw, subplotpars, and fig.subplots_adjust() behavior

Acceptance criteria:

• the skip marker is removed
• the test passes locally and in CI
• the assertions are specific enough that unrelated warnings do not cause failures