Turn on testForkGrouping by default

[lihaoyi]

Small projects probably won't notice the slight increase in overhead with testForkGrouping enabled, while larger projects would benefit substantially from the parallelism. And people can turn it off if necessary.

Mill's been using it since 0.12.0 and it's been pretty robust, probably can make it the default in 0.13.0

[lefou]

In my larger projects, testForkGrouping increased test time significantly. This is most likely project and hardware specific but I have yet to find the project where this setting results in relevant faster tests on my hardware.

[lihaoyi]

I think it depends on how big your test suite classes are and how long they take to run. If your classes are numerous and fast, then the overhead of spawning JVMs for each one dominates. If they are few and slow, then the overhead is small and the parallelization benefits win out

[lefou]

It looks to me this should be a runtime setting rather than build script defined, since it may vary for user / machine / use-case and also ideally should not affect any built outcome.