refactor ppo callback to move its logic to single controller (part 1)

[bowenyang008]

This is a series of efforts of moving controller logic from callback to a top level single controller, this part 1 does:

1. Update a couple ppo functions to fix bugs and make them static for sharing
2. Organize single controller to different actor groups
3. Refactor logic from callback to single controller
4. Create a simpler single controller callback just for dataloader and reward calculation (this should be moved to single controller as well in part 2)

[jdchang1]

LGTM so far

[rithwik-db]

yay

[abaheti95]

LGTM! Thanks!