TK: renderer performance optimizations (pass #1)

The renderer is now stable enough to optimize it. This means measuring the rendering loop's performance to find the most costly operations.

A lot can be done by inlining calculations in C functions and caching repeatedly used values. Also only updating changed sprite properties.

The goal is to have performance that's comparable or close to Cocos2D v3's tilemap renderer.

Estimation: 1-5 days (depending on how easily performance goals are satisfied)