This was the second attempt to encode run length data (first attempt used
huffman encoding).  The results were not as optimal huffman encoding would
provide, but the encoder/decoder is much simpler.

The data lengths were encoded using variable length numbers, similar to those
used in MIDI files.  Instead of 7 bits per group, the optimal bits per group
were determined emperically:

bits per group  dejiko  puchiko
6               6083    5229
5               5268    4560
4               4765    4050
3               4224    3844
2               4014    3558
1               4456    4000

Encoding assumed alternating run/space data, synchronized after newlines.
Thus every line must begin with at least one whitespace.

Second try, and I realized that I don't have the time and space to do what
I wanted in the beginning :P