- Instead of computing a complete score for each state, compute+compare
  the first component for each one, then the second, etc.
- Set colskip (and nrows) once and update occasionally once instead of
  constantly recomputing it.
- Add some more actions to do on the destination column.
- Fix move loop detection.
- Do not wait on bomb explosion (since they no longer cause loops).

Tweak strategy parameters, add delay on each move to avoid inaccurate state parsing, refactor some code