dataset_states_to_obs.py with camera depths

[WangYixuan12]

As discussed in Issue #98, I created a PR, which enables extracting depth from the demo. In addition, I added camera extrinsic and intrinsic information so that we could retrieve the point cloud. The aggregated point cloud is shown here.

Since I do not have a very deep understanding of robomimic, I am not 100% sure whether my implementation will break other things. So feel free to point out my mistakes!