Consider testing more optimization option variants

[dcodeIO]

Left the following comment on #1199 not directly affecting that PR, but still deserves an issue on its own :)

I still believe we should somehow test more variations of optimization options, just to make sure we are not missing a branch, since if something goes wrong on a user's end these things will be exceptionally hard to track down. Probably fine to not have fixtures for these, so just compile and run the tests on them.

Background is that there are some static branches in stdlib picking different code paths, like when optimizing for size or speed, that may become more complex than just default opts vs. no opts at all, and at some point we might end up not covering all of them in the test suite.

[dcodeIO]

Closing this issue as part of 2020 vacuum because we talked about this briefly, and it was mentioned that current static branches should align well with no optimizations and -O3. Let's keep an eye on this and re-open if necessary.