Implement an option to globally set/fix the number of parallel processing units

[xvallspl]

I would like to have a way to globally set/fix the number of processing units the parallel benchmaks will run on.

Some expert reviews of this potential feature:

"We already have the RB_TMP_DIR env variable, we could add another to control the argument passed to EnableImplicitMT" - @eguiraud

"we also have the rootbench datadir as a variable! (...) these vars can be easily exposed through cmake in the ADD_BENCHMARK_FOO functions" - @stwunsch

[vgvassilev]

Can you elaborate on the use case?