[WIP] Closures prototype

[nhuurre]

Add closures to the Stan language. The math library implementation is stan-dev/math#1924

See also the desing-doc stan-dev/design-docs#6 and the forum discussion https://discourse.mc-stan.org/t/closures-discussion/14932/

A closure definition begins functions keyword and otherwise looks just like a function definition. The keyword isn't ideal but something is needed to avoid ambiguity with ordinary variable declarations.

[rybern]

Hi Niko. You mentioned on Discourse that the parser isn't behaving. I just went through the gauntlet with the parser for declarations, so I thought I'd have a look.

The issue is a reduce/reduce conflict between return_type inside closure_def and sized_basic_type inside var_decl. You could %inline return_type, but then you run into the more fundamental issue of a reduce/reduce conflict between basic_type and sized_basic_type. Essentially, when the parser sees int..., it doesn't know whether to reduce the int to an unsized type (leading up to a declaration) or a sized type (leading up to a closure). It becomes unambiguous when the parentheses are read, but it needs to be unambiguous at the point of decision between sized and unsized.

I came up with a trick to solve a similar issue in #561 that I think I can apply here. But, since it'll take some work, I'd rather do it when you're stable here and when #561 is merged (since it included some refactoring).

[nhuurre]

Closing in favor of #742