Real world ball and cup environment

[nkumar-bdai]

This PR is a start on a real world BallAndCupStickyTable environment. Some notes:

1. Since placement needs to be super precise, we really want to get the position of the cup to be as accurate as possible. For this to happen, we need to observe the cup from one of the cameras right before we go to place (depth readings tend to be super unreliable from far away). The main thing this PR does is change around a bunch of the operators/predicates to enable this.
2. This env has different objects, etc. than the base env, so I copied the entire base map and made a new folder with a different metadata file. Probably redundant/wasteful, so maybe we'll want to do something nicer.

Rough edges with this PR/things to note:

1. currently, the robot always keeps a forward orientation when picking the ball; this isn't great
2. Sometimes, the position of the bowl is ~50cm off, because of a faulty depth reading from far away. This makes the robot believe the bowl isn't on the table, and that returns a planning failure issue (since the bowl is neither on the floor, nor on the table). I fixed this by making the table HUGE, but this is definitely a hack
3. The robot sometimes fails to see the ball inside the bowl when it succeeds: probably will need to adjust the pose of the final look action.

Next steps:

1. Figure out how to pick up the bowl after it falls on the floor and reliably put it back on the table
2. Get a reasonably-working oracle
3. Modify domain so that planning graph is fully closed
4. Figure out how we want to parameterize all our samplers
5. Start running experiments, and deal with deal with all rough edges that prevent us from getting good results
6. Profit

[tsilver-bdai]

LGTM! Overall, I'm pleased with the relatively small number of changes, considering that it's a whole new real robot environment. Great work!