`cabal-validate`: Steps are not granular enough, add groups of steps

[9999years]

Each --step in cabal-validate runs multiple test suites.

This makes it difficult to pinpoint which test suite caused a failure. This also makes rerunning tests (while fixing a failure) take longer than it needs to, because you may need to wait for unrelated test suites to finish. (The --pattern option usually helps, though.)

Proposal: Make steps run a single test suite instead of multiple, and introduce 'groups' to run multiple steps.

It might be nice if groups/steps have a nesting hierarchy, e.g. build builds everything, build-lib builds Cabal, Cabal-hooks, etc., and build-Cabal builds Cabal only.