Statistics use constant arrays instead of ML predictions.

[mjtb49]

Currently, PlayerPathTracker treats every room as if the policy was a constant array, and computes difficulty as if every room had weight 0.5.