Several issues were detected during rr playback of a recent assertion failure:

1. The balancing of generations that happens at the end of free set rebuild is redundant with the actions taken by ShenandoahGenerationalHeap::balance_generations(). This sometimes results in performing these actions twice, resulting in an unexpected imbalance of generation sizes and a generation size assertion error.

2. The convention within ShennadoahFreeSet's ShenandoahRegionPartition is to use rightmost equal to -1 and leftmost equal to max to indicate an empty partition. At the end of ShenandoahFreeSet::find_regions_with_alloc_capacity(), we were accidentally violating this convention by using righmost equal to 0 for empty.

3. At the end of ShenandoahFreeSet::reserve_regions(), we log failure to reserve intended amounts. But the test for when to log this failure is wrong. We were testing for capacity rather than availability within the relevant partitions