[gnugo-devel] DFA shuffle.

[Trevor Morris]

http://www.public32.com/games/go/trevor_3_13.1

- dfa_rt_t.next now stores offset to next state.
- shuffles DFA states to minimize jump size.
- shorten generated pattern files to avoid too many lines warning.

The main reason to change the DFA code to use incremental offsets
is that it allows for bigger DFAs.  The current code breaks when
compiling owl_attackpat with "-a".  This patch solves that problem.

Performance-wise there is almost no difference - someone else should
probably re-test that there is no performance penalty under GNU/Linux.

Conflict slightly with Paul's recent patch.  Either Paul's e-mail program
wrapped lines or mine did incoming, so I didn't try to reconcile it.

-Trevor