GC improvements

[purbon]

Usually GC might be an impediment for proper benchmarking, we've to be sure the final numbers are not too biased by this.
Might be we can start a GC before each execution, but don't add this to the timing.

[colinsurprenant]

GC activity is "normal" and I think it might be interesting to try to spot abnormal GC signatures or potential memory leaks for example. I like to run VisualVM once in a while to visualize GC activity and check for "unhealthy" activity.

GC might distort benchmarks on very short tests but when running longer tests, GC activity is just normal and part of the run time.

[purbon]

@colinsurprenant sure we can workaround GC affectation with long running test, but at the end is a way to average/normalize its influence.

One thing that we might do is running the test multiple times, remove the outliers and take the average, usually you can get pretty nice approximations like this.