Better Vertex caching

[emmauss]

We can improve the vertex caching of Pica, by:

• Using a better caching data structure, having O(1) access time.
• Caching the output vertices, instead of the output registers

Consequence of the patch:

• Increases memory footprint slightly, since we increase
  the size of the vertex cache.
• Performance increase, since more values are cached.
• Performance increase, since cache lookup time is now O(1).

From my performance analysis of Pokemon OR, I see that:

• ToVertex is a big hotspot (7% during intro scene)
• WritePicaReg is also a big hotspot (5% during the intro scene)