[WIP] Sema: Resolve type witnesses for conditional conformances early

[AnthonyLatsis]

Resolves SR-7516.

Allows for correctly diagnosing references to default type witnesses with
unsatisfied requirements once we start realizing types in members.

This change also establishes a definite type witness resolution order when associated type collisions occur in the presence of multiple conformances with differing conditional requirements. Previously, depending on validation order, a default type witness for a conditional conformance could satisfy a type requirement for a regular conformance, leading to miscompiles like the following:

 protocol A { associatedtype X }
 protocol B { associatedtype X = Bool }

 struct Foo<T>: A {} // no objections, A.X := Bool
 extension Foo: B where T == Int {}

Given the conditional requirements of a conformance are satisfied by those
of another conformance, default type witnesses for the former, more general
conformance, can now satisfy type requirements for the latter – but not vice versa – in
case of associated type collisions.
For example, both conformances in the snippets below will associate X with a single type witness
synthesized for the more general conformance to B.

 struct Foo<T>: B {
   internal typealias X = Bool // synthesized!
 }
 extension Foo: A where T == Int {}

 struct Foo<T> {}
 extension Foo: B where T: Equatable {
   internal typealias X = Bool // synthesized!
 }
 extension Foo: A where T == Int {}

[shahmishal]

Please update the base branch to main by Oct 5th otherwise the pull request will be closed automatically.

• How to change the base branch: (Link)
• More detail about the branch update: (Link)