decrease ramp-up time to <= 10ms

[rkeytacked]

With the switch from version 1.0.* (isbn-core) to 1.1.* (fast-isbn) the parsing speed tripled (from 4M ops/s to 13M ops/s). But the ramp-up time increased from 10ms to 30ms. This is due to parsing the the compact decision-tree data and creating the 13-dimensional character-lookup-tree.

We need to find a way to have the character-lookup-tree faster at startup, but without increasing the byte code too much.
A good trade-off between for the amount of generated code has to be found.