Revert "Revert "Test whether OSX Travis can go back to running all tests""

[tkelman]

Reverts #15782

might fix? #15808 by actually running real tests instead of just compile-all which times out.

edit: doing all tests also looks borderline for timing out, but maybe #16012 will get things back down

[tkelman]

I've been restarting the mac build here a bunch of times and it's been passing pretty reliably now with all of the normal tests running, but not doing compile-all testing. If it passes again I'm temped to merge and see how well we can keep within the time limit. nevermind it didn't

@vtjnash this conflicts a bit with your travis changes from #16059, but as mentioned in #15726 (comment) I think it would be most appropriate to do compile-all testing on a separate, non-travis service. Could you respond to

There are competitors to Travis that we can turn on, with similar yml configuration, many of them are docker-based so we can put up and pull from a public docker container build that has the deps prebuilt. Would it be okay to have the --compile=all testing only run on master commits and not PR's, if that made setting it up easier?

Depending on whether you want to run compile-all testing on every PR or not, that determines which Travis-alternative services we can look at.

[tkelman]

I've been restarting this a bunch with the subarray tests skipped and it's been working consistently. All but subarray is better coverage (and hopefully will time out less often) than what we're currently doing with a minimal set of tests plus the timing-out compile-all.

I think the makefile changes from #16059 are uncontroversial, but the CI changes would be better moved to a different service that only does compile-all testing.