Shenandoah: pre-Full GC root updates may miss some roots

Look at Full GC marking code:

void ShenandoahMarkCompact::phase1_mark_heap() {
  ...
  cm->update_roots(ShenandoahPhaseTimings::full_gc_update_roots);
  cm->mark_roots(ShenandoahPhaseTimings::full_gc_scan_roots);
}

It enters update_roots, which ends up calling this:

void ShenandoahRootUpdater::roots_do(uint worker_id, IsAlive* is_alive, KeepAlive* keep_alive) {
  ...
  _serial_weak_roots.weak_oops_do(is_alive, keep_alive, worker_id);
  ...
  _weak_roots.weak_oops_do(is_alive, keep_alive, worker_id);
  ...
  _dedup_roots.oops_do(is_alive, keep_alive, worker_id);
}

But the problem is, at this point we already have clear bitmap, so "is_alive" always replies "false", and so the particular roots might not be updated at all. This is an obscure corner case that matters when heap has forwarded objects in roots that need updates, and we run into Full GC.