For heap root merging we size the number of threads based on the number of collection set groups to be evacuated one-to-one.

This may understate the need for cpu as the collection set groups to be evacuate may have a wildly different number of cards to be merged.

(This has recently been introduced/changed with JDK-8342382, from number of regions in the optional collection set to number of groups for optional evacuation. Since optional groups are mostly of size one or at least very small, this does not make a big difference, so it is not a regression or such)