Grasp gen script update

[alberthli]

Updates the grasp generation script (for one scale) with the following changes:

1. No train/test/val split. All images are saved to a directory called images, and this change is also reflected in a single transforms.json.
2. All other outputs of this script are not generated except optionally segmentation/depth images (false by default), which you can generate by passing in args.
3. The scale of the images is saved up to 4 decimal places of precision in the file name.

TODO:

• scale the camera positions with the object scale