format:
 Experiment with putting all keys before all data
 can load factor be more scientific?

sdb_make:
 linked list -> linked list of arrays
 allocate multiple skunkrec_t at once
 use iovecs?
 speed up the parser somehow..
 speed up collision resolution
 internal prime-number table for optimum table sizes?

sdb_fsck:
 consistency checker program

sdb_stats:
 stats program giving breakdown of distances
 point out hash collisions
 work out how evenly the data is distributed in hash tables

sdb_query:
 support cursors

api:
 support cursors

python module:
 use exceptions
 make query object a dictionary
