Refine plan demo

[ct2034]

Replication of convince-project/data-model#28

[ct2034]

Description by @charlie1329

Introduces the REFINE-PLAN part of the overarching demo.

Here, the robot is looking for a new object (bread) in the environment. After learning from data, the robot learns to search for it anti-clockwise through the environment.

Adds a new package, refine_plan_demo, in the simulation directory. There is a README in the package on how to run it (this can be translated directly to the web page).

This PR makes the following changes to the main sim:

Adds new dependencies to Dockerfile/requirements.txt
Changes the Pyrobosim commit to support probabilistic object spawning
Changes the motion planner to A*. I was observing issues under RRT* where the robot was frequently taking very suboptimal paths between locations. This would render REFINE-PLAN unable to do anything useful.
Moved two tables slightly further apart as otherwise the navigation (regardless of planner) was unable to navigate between the two.
Added an override in run.py to set the detect success probability to 1.0 (we can't handle partial observability in REFINE-PLAN).
If anything isn't where you'd want it to be, let me know.