Make seed independent of num.threads and add legacy option

[erikcs]

This PR makes the random seed argument produce the same results irrespective of what num.threads is set to. In order to reproduce past results, a global package option options(grf.legacy.seed) is added which can be set to TRUE to revert back to the old behavior where seed and num.threads interacts (see grf_options for an example).

#1368.

[jtibshirani]

Great we're finally fixing this. I do wonder if the grf.legacy.seed option adds more complexity than it's worth ... it's pretty easy to install old versions of R packages, and for reproducibility it's good practice to save and fix the package version. And now, we'll have to figure out when it's save to deprecate and remove the new option grf.legacy.seed !

[erikcs]

Yes it adds complexity, but that's the price we have to pay in order to do this imo! Past versions of GRF are not available as CRAN binaries and have to be installed from source, which unfortunately can be challenging for many users.

(there's a bunch of unrelated commits below because I got a new arm mac that evidently produces slightly different results for characterization tests stored on the repo using x86-based systems)