GitLab BIG_TESTS hack breaks CI_TESTS

[Dan Bonachea]

The summit/pgi gitlabs are partially ignoring CI_TESTS as a result of the BIG_TESTS hack.

Eg: https://gitlab-ci.alcf.anl.gov/anl/upcpp/-/jobs/51554/raw

As a result the pipeline runs unintended tests, which (as in this example) can lead to spurious failure of overall pipeline if the experiment is specific to a particular test set and not intended to be appropriate for other tests.

Ideally the script would perform an intersection, but failing that a better solution when CI_TESTS is set would be to just disable parallel test execution and the BIG_TESTS hack

[Paul Hargrove]

I was over thinking the problem initially.

Basically if CI_TESTS and BIG_TESTS are both set, we set the latter equal to the former to achieve both goals: tests are limited to those in CI_TESTS and (at least) those in BIG_TESTS are compiled sequentially. The loss in parallelism for a presumed small set of tests in CI_TESTS should be acceptable.

Fixed, as can be seen in the following retry of the job in the original report

https://gitlab-ci.alcf.anl.gov/anl/upcpp/-/jobs/51568/raw