Test the "launch" functionality using GPU and vectorized CPP

[valassi]

Test the "launch" functionality using GPU and vectorized CPP.

This is point 5 in #576.

This can be done at two levels.

Eventually, when the implementation of code generaion is fully integrated upstream, one can run "generate", "output" and then "launch". This is clearly our final, golden, integration test.

But already using the present gridpacks (see Stefan's #620), where "generate" and "output" are pre-cooked, it would be useful to understand what happens when one types "launch". Amongst other things, try to understand if the parameters in madevent are ok for vectorized CPP (need at least ~32 events per generation) and especially the GPU (need ~10k+ events per generation...).