Add image observations/improve render for BlackBoxWrapper

[ottofabian]

Currently, the render function only supports "human" format and does not return anything for the array case. This should be improved to used Black-Box envs with vision obervations.