Build mkgmap/splitter #209

- add step to do a final optimization. When the global optimization finishes with a result where tiles have less than 80 percent of the max-nodes then try to optimize the partial splits. - accept solutions which have less than 3 percent of the tiles below 33 percent fill ratio as nice. Might help to find good solutions for input with lots of empty space (sea or deserts). - use Long instance for minNodes to reduce memory footprint as the same value is frequently added to the HashMap - don't trim tile before it is actually needed for testing a split - improve speed of TestGenerator a bit - improve again the method to restart after searchLimit was reached. Now the exact position is restored and thus no tests are skipped or performed twice. - allow "SOME" method when --num-tiles is used This again improves both speed and split result where older version endet with a split where the worst tile fill ratio was below 80 percent.