Split the integration tests into "fast" and "slow" sets

[odisseus]

Some integration tests are decidedly slower than others. As far as I know, slowness is mostly caused by compilation (especially if external layers are involved) and also by importing of said layers, but the certain way to check that is inspecting the timestamps in the CI logs.

I think the tests should be split into two sets: the fast ones (those that take less than 10 seconds in CI) and the slow ones. The CI runner should start the slow set after the fast set passes.

There should also be a single command to run all the sets (fast, slow, and pending) in succession.