Ideas for improving the performance of fsck:

(1)  Utilize a different mechanism for storing the (struct direct)
     entries throughout phase2.c, to remove implicit memory intensive
     redundancy.

(2)  Encode the number of links per i-node count differently, in order
     to save space taking advantage of the typical usage under Coherent.

(3)  Arrange space for the tables in a total needed way to avoid later
     fragmentation problems affecting the memory requests for successive
     file systems.

(4)  Devise a more clever buffering scheme for the blocks containing
     i-nodes to understand the differences between phase1, phase2, and
     phase4.
