Currently all collectors except G1 implement the GCOverheadLimit functionality, i.e. G1 will not exit the VM if the amount of GC activity is larger than a given CPU usage with an OOME.

Implement that similarly to other collectors; make sure that unlike Serial and CMS (but like Parallel) GCs caused by TLAB allocations are included in the calculation.

It is easiest to do that by not causing any GC when trying to reload TLABs (like Parallel)