Add support for chemical heuristics into reward functions/allowed moves

[WardLT]

@nkkchem mentions that there are certain rules that molecules must satisfy in order to be valid.
For example, "presence of an epoxy" group is an example. We need to make sure that we can either restrict the agent from violating "hard" rules or impose "soft" constraints by adding in penalties to the reward function.

[WardLT]

More rules that we should include:

• Do not have many benzene rings
• Should have at least one carboxylic group