Fix binding- and assignment-expression related typing issues.

Simpler solutions are better, so the typechecker is back to normal.
Instead we simply make sure that assignment expression is never
nested under another expression, and binding expressions may only
be nested under LogicalAnd-expressions. If only I knew why I thought
type shenanigans were a good idea in the first place...

Put expression types of procedures in type table

Before we annotated the AST. But that is silly in hindsight:
certain operations have a completely different meaning when applied
in a polymorphic context (i.e. the field index determined by a
select expression).

Rewrote typing to use indices.

Currently it is slower than before, because we do a HashMap lookup
followed up by actually using the index. But it serves as the basis
for a faster type inferencer.

The main goal, however, is to fix the manner in which polymorph
types are determined. The typing queue of functions still needs to
correctly write this data to the type table.