Fix type system logic for detecting if a type param can match itself.

[jemc]

This PR fixes type system logic for detecting if a type param can match itself.

For example, this snippet doesn't compile with current ponyc master, but should:

trait val T[A: T[A]]

class val C1 is T[C1]
class val C2 is T[C2]

primitive Foo[A: (T[A] val & (C1 | C2)) = U64]
  fun apply(a: (A | None)): Bool =>
    match a
    | let _: A => true
    else false
    end

actor Main
  new create(env: Env) =>
    env.out.print(Foo[C1](C1).string())
    env.out.print(Foo[C1](None).string())

Note that this breaks a few methods in Iter that are actually capability-unsafe, so we have to figure out what to do about that. I'm marking as "DO NOT MERGE" and "needs discussion during sync" so we can discuss this aspect.

This may be related to #2584, but I don't think it has the same problem as @Praetonus' solution in #2597. Note that my code example above doesn't have the issue of None being a possible subtype of A. As such, I don't think we have to wait for @Praetonus' RFC to merge this. Still, I'll ask @Praetonus to comment here if he can.

This may also be related to #2646, which had a runtime segfault in Iter.filter_stateful, one of the methods found to be unsafe by this type system fix.

[Theodus]

The filter_stateful, filter_map_stateful, filter, filter_map, skip_while, and take_while methods will need to have their return types changed from Iter[X]^ to Iter[X!]^. The current return types are unsafe and impossible to implement correctly on a generic Iter[A], since a predicate {(A): B} cannot be given the result of Iter[A].next() and the result of iter[A].next() cannot be stored without becoming A!.

[SeanTAllen]

@jemc the example given now compiles with ponyc master. Do we still need this? Should I spend time getting this capable of passing CI and not have conflicts? Or should we close it?

[SeanTAllen]

@jemc and I discussed, I will open a PR with just the regression test in test/libponyc/matchtype.cc added. If it passes CI as expected, then we will merge the new PR and close this. Otherwise, Joe and I regroup.