Fix calculation of G1 effective scanned cards prediction

When calculating base time needed for evacuation, G1 tries to predict the cards that are actually scanned during the next gc. This calculation is bugged:

  size_t effective_scanned_cards = _analytics->predict_scan_card_num(rs_length, collector_state()->in_young_only_phase());

The calculation completely ignores the cards pending in the log buffers which often make up a significant part of the prediction.