Split hash buckets from the unique node table values

[SSoelvsten]

As far as I can tell the hash value in each node really is just a separate array for all buckets. Assuming it really is only used in bdd_makenode then we should be able to get the following new properties by splitting it.

• This decreases the node struct size to be only 20 bytes, which may improve cache locality (though it still does not nicely fit into a 64 byte cache line). Combining this change with Change nodetable and cache indexing from int to ssize_t  #3 we get a node size of 32 bytes, which fits perfectly with the cache line.
  • Recursing through the BDD may run faster this way due to better cache behaviour.
  • One can hope that a lookup of the hash bucket leads to fewer nodes being thrown out of the cache, since merely another cache line with a hash value is thrown out instead. I find it unlikely, but maybe the cache behaviour of the hash function improves.
• This allows the node table to not have to be a prime number in size and so make use of the entire addressable space.