Avoid sending per-region GCPhaseParallel JFR events in G1ScanCollectionSetRegionClosure

In G1ScanCollectionSetRegionClosure::do_heap_region(...) we send JFR events on a per-region basis which lead to a lot of events. The reason we don't send the event for the whole iteration is because the closure includes two phases: OptScanHR and CodeRoots/OptCodeRoots.

We can solve this in at least two different ways, we could add a new phase that is named to signal it does both things or we could create a new event that can include two different durations for the two included phases.