Revert prior attempted optimisation. Gridding the land pays in some situations, but not all. Restricting to an upper bound might help, but overall, seems too fuzzy to be worth it. On one side is increased cost of Add/Delete + extra test on collision check, on the other is skipping the list iteration. Perhaps for large lists.

IDI_ICON1	ICON	DISCARDABLE	"hedgewars.ico"