mode7 speed fixes:
	+ Don't draw sky every frame, only if needed

	+ faster multiply routine
	+ If result is AABBCCDD we only need BBCC for fixed point
		result
	+ Pass NUM1H in A? no because used multiple places
	+ inline hlin_setup? (save 12 cycles)

	+ re-arrange variables to better take advantage of self-modifying code
	+ only doing the spacez calculation if it has changed
	+ update the constants to be constants
	+ move the screen width constant to own set of varaibles
		(instead of temp)
	+ leave one of multiply results in accumulator at end?

	+  Skip key parsing if no key read

	slower: move to 40x40 again
		draw every other line.  First with color/black		
		then xor in the second line

short term:

longer term:
	sky -- color looks too much like grass on monochrome display
	map -- adjust x/y to better match real position
	name entry screen
	inventory screen
	collision detection
	after game over, return to title screen
	save/restore game