Create TRS search methods that capture the behavior of less flexible formalisms

[joshrule]

Many slightly less-flexible formalisms can be approximated with a suitably restricted TRS learning algorithm. These include:

• combinatory logic (learn rules that don't use variables)
• LOT-style PCFG (learn rules of the form f(x1,x2,...) = <definition in terms of existing operators>)
• lambda calculus (learn PCFG-style rules and rules which include only variables on the RHS)

It would also be great to setup classic priors for some of these models, including Rational Rules (Goodman, et al., 2008).