Base IR.typ_bound on con, not var.

[nomeata]

Previously, IR.typ_bound would mirror Syntax.typ_bound, with a var
instead of a con. But post-typechecker, we only want to work with the
stamped con. Previously, this was hidden away in the annotations
(misusing Type.typ as `Con.t option).

This patch cleans this up a bit,
but introducting a Type.open_bind that binds a Con.t to its bound,
and using that as part of FuncD.

[nomeata]

More context (copied from Slack)

• In source types (Syntax.typ), everything is simply named (id, i.e. string), including Syntax.typ_bind.
• In internal types (Type.typ), the type parameters of a Func are always represented as de-bruijn-indexed variables (Var of (string, int), but the string is just for printing), The Type.typ_bind has a name, but it is again just for printing; the position is important.
• Type annotations in post-type-checked source use con in a named way. For example, when type-checking id<A>(x:A) : A { x }, the type checker creates a fresh con for A, and the type annotation of the VarE "x" is Con(c,[]). It records the fresh con in an annotation to the typ_bind (abusing Type.typ as a con option).
• The IR uses Type.bind for the typ_bind in FuncD, but it is not a great match, because Type.bind is meant for de-Brujin-indices, and the name field of it is useless. Therefore, the con still needs to be put in an annotation, and extracted from there.

I tried cleaning up the last point, by introducing Type.open_bind that has a con and its bound. See #159.

[crusso]

LGTM

[crusso]

Really LGTM