possible bug in multi-agent example

[NansDarraillan]

in examples/maze_multiagent_mdp_multisolve.py, in the _state_sample method, for each direction, in case the agent can proceed in that direction and next_state_1 is updated, I think the second next_state_2 update should instead be an update of next_state_3. Otherwise next_state_3 is always state, and one of the 2 low probability directions is always favoured against the other if both are free.

I opened it just reading the source, please close this issue if this is the expected behaviour.