Clarify computation model for Pikov animations

[tswast]

We should document the intended computation model for Pikov animations.

Thoughts:

• It's a state machine (with stochastic transitions instead of based on deterministic rules).
• Except it has extra state like the animation's current position, in addition to the current state.

Question: Is there always 1 image per state? (For now, yes.) Maybe it makes sense to track direction separately, for example?