Nested Learning: Two-State World & Observer

This demo shows a minimal nested-learning setup:

• Level 0: A two-state Markov world with fixed switch probability P_CHANGE.
• Level 1: An observer that learns a transition matrix M to predict the world's next state.

The observer updates its internal model using prediction error between its probabilistic forecast and the actually observed next state.

This is a tiny illustration of nested learning:
a learning process about another process.