After JDK-8192647, a GC-request, once inside a safepoint, will run-to-completion. Therefore, callers who requesting GCs don't need to retry in a loop any more.

For example, the loop `ParallelScavengeHeap::collect` is unnecessary.

```
  while (true) {
    VM_ParallelGCCollect op(gc_count, full_gc_count, cause);
    VMThread::execute(&op);

    if (!GCCause::is_explicit_full_gc(cause)) {
      return;
    }

    {
      MutexLocker ml(Heap_lock);
      if (full_gc_count != total_full_collections()) {
        return;
      }
    }
  }
```