NumPy -> Torch

[hang-yin]

High level changes:

• Numpy -> Torch conversion
• Enforce th.float32 whenever possible
• Simulation backend -> torch
• Rewrite transform utils with torch jit scripts
• Use math instead of torch whenever possible
• Fix robot state tests
• Remove Scipy rotation library
• Remove trimesh transformation library
• Reenable examples as tests
• Fix all examples (except attachment demo seg fault)

TODO:

• [x] Bring performance back by getting rid of warp.synchronize() call when fetching data from annotators

[ChengshuLi]

Really great work Hang! A couple high-level comments:

1. You mentioned a performence regression due to synchronization. Is that resolved?
2. For each of the issues I mentioned above, you should consider doing a global search, because I am not being exhaustive for every occurrence of these issues.
3. Can we make sure all the tests / examples as tests pass before merging this PR? What's the current status?
4. Currently I rubberstamped the transform_utils.py because I am not sure whether the math is completely correct. Any suggestoin for how I should actually review this?