Enumerate by size, not depth

[ninehusky]

Relates to #15.

Had a conversation with Loris today about this!

Right now, the enumeration algorithm I've implemented isn't enumerating by size -- instead, it's applying each production to the entire corpus of programs generated so far. I think in the beginning, this is creating programs that look like they're being enumerated by depth, but this won't scale.

[ninehusky]

Closed by #19.