swift234

[AST/Sema] Typealias in protocol further corner fixes.

Don’t try to find a conformation witness for typealias declarations in
protocols.

Fixed a bug with same type constraint between an alias and a concrete
type - just a bad assumption on my part. Added test for that.

Made archetype builder more resilient in constructing PA trees: it will
work correctly now if the alias destination is a dependent member type
instead of an archetype type. It also handles the case of finding
multiple aliases with the same name along with an associatedtype with
that name, fixing up all the representative ptrs.

Extensions/Nominals that tried to use a protocol's typealias would get a
dependent type as resolved with the protocol's base instead of with the 'Self'
base type of the current extension/nominal, resulting in spurious conformance
failures. So resolve aliases to protocol assoctypes based on the 'Self'
in which they are used.

Also fixed tests to not use common stdlib names, added tests for
typealias in protocol extension and self & recursive aliases. One recent
crasher also fixed.