Run example runcards in tests

[Zaharid]

It would be good to run some of the example runcards in tests. There is the question of whether that should be done always on every commit or e.g. like with the fit running when we set some label.

In any case it should be done reasonably (e.g. downloading all the nine theories for theory covmats is out of the question) and reusing resources as much as possible.

[scarlehoff]

Skipping the future tests and maybe the ones with 1000 replicas (and being more carefuly with the reuse of resources) one can actually run al tests in not much more than 30 minutes. So it could even be done in parallel with the usual tests without any loss of performance for us (Microsoft can absorb the hit, I'm sure).

The question would be, what makes the test "pass". Do we just want to see that they pass? Do we use them as regression tests? Should they just be accessible for the reviewer of the PR to check that nothing broke (and how would they even check, regression test by eye?)

What about a label that will run it for both master and the PR and comment with the changes, and then the reviewer / user can decide whether it is an acceptable change or not (the baseline being always master)? This would take about 1-2 hours.