ci: Improve handling of pipeline errors from flakes

There's an issue with listing flaky tests in the skips lists as we've been doing: If the flake is due to some bad state left from the previous test, then adding to the skip list my just move the flaky result to the next test to be executed. This has been an issue with some of the freedreno flakes people have been seeing which I hadn't banned.

Fix this by post-processing the unexpected results and cleaning out expected flakes, letting the pipeline pass (but still reporting them as normal). This makes flake management easier, and also gives us logging so we can see when a flake has cleared up and we can drop it from the list.