RFC: Enable threading build by default, but with only 1 thread

[tkelman]

unless you opt in to setting JULIA_NUM_THREADS
closes #15743

Anything else I missed here? Let's see what kind of regressions this might bring:
runbenchmarks(ALL, vs = "JuliaLang/julia:master")

[nanosoldier]

Your benchmark job has completed - possible performance regressions were detected. A full report can be found here. cc @jrevels

[jrevels]

As you're probably aware by now, the regressions here (and likely the improvements as well) come from very noisy benchmarks, especially sort. It shouldn't be long until my fixes for all this stuff get officially deployed to @nanosoldier...

[StefanKarpinski]

Are we using a fixed seed for the sort benchmark?

[jrevels]

Yes. The issue is that the old version of that benchmark (the one currently on @nanosoldier) naively incorporates the allocation of the input vector into the sample measurements. This is fixed in the new version of the suite, but I'm still running tests on the new version and haven't deployed it yet.

[tkelman]

Sounds like we should go ahead and do this then?