Enable testing a particular problem and/or executing a particular random run

[zaikunzhang]

Motivation:

Errors / weird things may occur when testing a particular problem or executing a particular random run. We should be able to reproduce them quickly, not rerunning all the tests but running only a minimal subset of them.

N.B.:

1. If randomization is involved, make sure that the random seed is reproduced.
2. If the whole test runs in prallel, then the subset of the tests should also run in parallel; otherwise, things may not be reproducible --- you should know this and you should know why.

Thanks.