Investigate scaling workers per phase in G1 Full GC

Currently the number of workers used by the G1 Full GC is calculated at the start of the GC. The same number of workers might not be optimal for all different phases and we should investigate if this can be improved.

An example is clearing of the bitmaps which is done both before and after the GC. For this type of work using as many threads as possible is often good, while other parts might need less workers. Right now this is solved, by always using as many workers as already created for the clearing, but if the workers for the full gc were calculated more than once this could also be included.