[5.7][Sema] Use ExistentialType for Any and AnyObject.

[hborla]

Cherry-pick of #41909

Note that the main PR is not merged yet because passing lldb tests depends on an lldb change that has to land on llvm-project:swift/release/5.7 first.

• Explanation: This change models existential Any and AnyObject using ExistentialType, just like all other existential types. Previously, these two existential types still were modeled with ProtocolCompositionType because they're not required to be written with any, but this allowed using any on these types in conformance constraint context, e.g.

func generic<T>(value: T) where T: any Any { ... } // okay??

This also caused type resolution to not know the difference between (any Any).Type and any Any.Type, as described in #58367.

Note that Any and AnyObject still do not print with any in the ASTPrinter, except when it's needed to distinguish between the existential metatype and the singleton metatype, e.g. any Any.Type vs (any Any).Type.

• Scope: This affects the frontend representation of code using Any and AnyObject as existential types.
• Risk: Low.
• Testing: Passed CI tests and the source compatibility suite. The ExistentialType representation itself has already been shipped in Swift 5.6 and thoroughly tested.
• Reviewer: @slavapestov

Resolves: rdar://94370283, #58367

[hborla]

swiftlang/llvm-project#4863

@swift-ci please test

[hborla]

@swift-ci please test source compatibility

[hborla]

swiftlang/llvm-project#4863

@swift-ci please test

[hborla]

swiftlang/llvm-project#4882

@swift-ci please test