Add egraph-unary for maximal unary context extraction

[pavpanchekha]

Unary expressions are special in numerics: they can be efficiently and accurately approximated by polynomials. So one thing we might want is to rewrite an expression like log(exp(a) + exp(b)) into a form like a + log(1 + exp(b - 1)), where there's a nice unary subcomponent like log(1 + exp(x)). Since this requires some rewriting, we want to do it on an e-graph.

This PR thus adds egraph-unary, which extracts non-trivial (not x or sin(x)) distinct unary expressions in an e-graph. It works is a pretty straightforward way, though I got here in a complicated way involving dominator trees:

• Each eclass e is a unary function of e with expression x
• Each enode f(e_1, ..., e_n) is also a unary function of e if each e_i is a unary function of e, with the obvious witness term

So you compute this via an eclass analysis where the value is a map { eclass → expr } such that if the data for e1 has { e2 → E[x] } then e1 = E[e2]. Pretty neat and pretty fast! Kinda ugly to do this on the Racket side but whatever.

[obround]

Looks very cool, I've been playing around with egraph-unary for interval-based series, and the expressions returned seem pretty useful.