The problem was once again a poorly written hash function, which caused ATN config sets to
hash poorly. The poor grammar caused many thousands of DFA states to hash to the same
bucket. In your average grammar, this situation would not arise.

For now, I have changed that hash function to use the memory address of the slice that
contains the configs, though i think that this is sub-optimal.  I will revist this as part of a
general attack at further performance gains.

Signed-off-by: Jim.Idle <jimi@gatherstars.com>

Signed-off-by: Jim.Idle <jimi@gatherstars.com>