brute force solver showcase.  i built the solver to look for a 20, but it couldn't find one, so instead i spent a while feeding it different layouts to see what it would do.  the way the solver models state transitions does not include drops, grabs, and swings that reposition the arm between drops and grabs, so these movements could probably be improved with more accurate modeling.  it also won't end a cycle with an atom on a hex that doesn't have a glyph on it -- hence the equilibriums.  for this layout, i also manually limited the max atom count to 9 on the board at once and allowed atoms to swing over hexes outside the glyph footprint.  the solver spent 20 seconds looking at 41,474,285 total board states to find the state where both outputs are dropped.