Remove Union Flattening

[KabirSamsi]

Perhaps this merits further discussion, but having gone through this a few times, I can see no genuine advantage of making the set type recursive. We currently have

type set =
  | Class of clss
  | Union of set list

I propose it to be rewritten as

type set =
  | Class of clss
  | Union of clss list

That is to say, Union([Class 'A'; Union([Class 'B'; Class 'C'])]) should be treated as equivalent to Union([Class 'A'; Class 'B'; Class 'C']). The latter is much easier and cleaner to work with.

Part of my work this semester involved formally proving that dequeue-side semantics for flattened unions behave the same as recursive ones (i.e., showing that the two sets I described above behave the same way). So I'd feel confident in removing that.

[polybeandip]

To clarify, does this means you want to change Rio's syntax to disallow unions containing unions? So, this program would be illegal:

classes A, B, C;

policy = fifo[union[union[A, B], C]];

return policy

and would have to be written as

classes A, B, C;

policy = fifo[union[A, B, C]];

return policy

If so, then I'm wondering if you could say more about why we should add this restriction? If not and only the AST representation (not Rio's syntax) should be changed, then perhap #76's tweaks to Policy.t satisfy the issue? The PR makes it

type t =
  | Fifo of Ast.clss list
  | EDF of Ast.clss list
  | RoundRobin of t list
  | Strict of t list

Ast.stream is still unchanged, but Policy.t'sset2stream policies do hold a "flattened union" now