I implemented the generic coder for JBIG2, first without using TPGDON,
and then later with it.  It seems that the compression is slightly
worse (about 1%) using TPGDON, though it might be faster.

(TPGDON uses one bit per row to indicate that row is a duplicate
of the previous row.)

I wonder if anyone else has noticed this?  Maybe I am not so surprised,
since the duplicate rows are likely all white, and the coder does
very will with them anyway.  It is possible that my code is wrong, but
not so likely.  A very large fraction of code errors result in
random bits out of the decoder.

-- glen