Optimized sprite rotation routines

[ZERICO2005]

This is in its own commit due to the complexity of gfx_RotateScaleSprite and gfx_RotatedScaledSprite(Transparent)_NoClip.

The inner loop of gfx_RotatedScaledSpriteTransparent_NoClip is 8F faster when a pixel is drawn. I also believe that creating a clipped variant of this routine is possible with a constant performance penalty.

I have fixed most of the issues outlined in #612. There are a few limitations to note:

• The sprite will render incorrectly if the output is larger than 255x255. This is not worth fixing however, since a 256x256 sprite doesn't fit onto a 320x240 screen anyways.
• input sprites larger than 210x210 (255 * (sqrt(8) - 2) or ~211.2489) may experience texture wrapping. This should be harmless, as shown below with a 254x254 sprite