Use line-by-line hashing of input file for twolevel_modelbuild.sh to generate reproducable "new file names" instead of counting lines

[gdevenyi]

Using some kind of hash will allow us to re-use partial runs while removing files without changing the order.