Write a Probe Environment that tests a models ability to look at previous observations

[jbloomAus]

We currently have 5 probe environments for single timestep models and I'd like a prob environment to test if a model can learn:

1. to take the correct action as a function of a previous observation
2. to test that the model doesn't learn the correct action if it's outside the recurrence range or something (more speculative)

[jbloomAus]

This turns out to be hard because we get underflow with the prob environments on the RNN and need to edit the trajectory LSTM model which expects to get mini-grid environment frames. I will need to think more about how to test it with probe environments.