Optimize the render loop

[sastraxi]

A few things were going wrong to make rendering a lot slower than it should have been:

• tint_mask turned into a general-purpose, caching helper for alpha painting
• a bunch of places refactored to use tint_mask
• rounded rects are now cached
• arc ring rewritten to be more efficient + cached
• remove font.font_metrics and instead cache get_text_size / get_text_bbox
• cache text drwaing
• add batching so immediate draws can't race coalesced draws
• coalesce new draws into pending pushes

As a result, the v2 is now much more usable and will have more CPU available for effects. However, the caches will use about 5-6 mb more RAM.

The snapshot changes are all small pixel alpha blending rounding changes; nothing visible has changed.

Finally, I removed all profiling code (was from a previous investigation and was never cleaned up).

[rreichenbach]

Looks good.