add warning to academy docs

docs/Learning-Environment-Design.md

 for each training episode. Otherwise, the agent would probably on learn to solve
 one, particular maze, not mazes in general.
 
+#### Academy Pitfalls
+
+In some games, it may be desirable to step the environment with the Academy manually via `Academy.Instance.EnvironmentStep()`.
+If this is the case, be sure `Academy.Instance.EnvironmentStep()` is not called directly or indirectly by the agent's
+`CollectObservations()` function as this will cause an infinite loop that prevents that main Academy update function from being called.
+
 ### Multiple Areas
 
 In many of the example environments, many copies of the training area are