Optimize graphx clipping logic

[calc84maniac]

Reworks sprite clipping to use range-based (unsigned) comparisons instead of signed comparisons, and also moves shared pointer calculation logic into the clipping routine. Additionally, logic is added to gfx_SetClipRegion to apply SMC to all clipping routines. This saves 178 215 bytes overall, and I measured the following changes in cycle counts for gfx_Sprite (up until the gfx_Wait for drawn sprites, or through the return for culled sprites):

fully onscreen: 1177 -> 716 644

clipped top: 1450 -> 860 772
clipped left: 1488 -> 864 780
clipped right: 1254 -> 756 684
clipped bottom: 1202 -> 744 672

clipped top left: 1761 -> 1008 908
clipped top right: 1527 -> 900 812
clipped bottom left: 1513 -> 892 808
clipped bottom right: 1279 -> 784 712

clipped left+right: 1565 -> 863 779
clipped top+bottom: 1475 -> 859 767
clipped top+left+right+bottom: 1863 -> 1006 902

culled top: 471 -> 370 334
culled left: 803 -> 540 468
culled right: 776 -> 540 468
culled bottom: 474 -> 370 334

I'll probably take a look at gfx_RLETSprite later and add it to this PR if I can improve it.

[calc84maniac]

While updating all clipped routines to use SMC, I found and fixed two bugs in flood fill clipping, the first being that X clipping on the left was always treated as 0, and the second being that X clipping on the right was off by one (treating the upper bound as inclusive instead of exclusive).