This needs a healthy dose of altering aftersom FPGA code changes, currently only works properly with 16x16 pointers.

I'm not quite sure if this ever happens, but something like this is required if we want to call draw_line from elswhere on the ARM core.

Replaces the big if, if else block with a switch/case jump table, which seems to result in a very big speedup for functions that are currently not CPU time limited on the ARM core.
(The diff is unfortunately a giant mess.)

This was a tricky one. When implemented, P96 replaces many (all?) of the BlitRect (`rect_copy`) that use a mask of 0xFF to allow for graphics minterm handling instead.
I suspect this is due to pretty much running out of registers when calling both of these functions.
The minterm handling currently isn't implemented, but it seems to work fine for all the test cases I have available.

Reusing register-accessible values for unrelated data ended up being a hassle while implementing Planar2Chunky support.
A good number of the other functions could/should probably be reworked to use these instead of reusing rect_x3/y3 and similar.

RTG acceleration improvements:
- Consolidates and speeds up FillRect implementations and Pattern Fills
- Adds accelerated DrawLine

Uses the Bresenham algorithm with a special case for solid color line drawing, which used to be the only thing actually accelerated by other RTG cards.
Line pattern support isn't 100% correct, but there is some commented out code in place for when it can be accessed through P96.

This is no huge speedup, but it does unroll the loops a little and keeps the actual bits intact, so it might be a easier to apply the mask.
(I couldn't actually figure out how the mask was supposed to work.)