[Basics] Switch IdentifiableSet to use OrderedDictionary

[xedin]

Motivation:

Turns IdentifiableSet into an ordered collection which should cut on flakiness in tests and other spots that expect certain ordering of elements.

Modifications:

Changes storage of IdentifiableSet from Dictionary to OrderedDictionary.

Result:

Less flakiness throughout.

[xedin]

@swift-ci please test

[MaxDesiatov]

Thanks!

[MaxDesiatov]

IMO definitely worth cherry-picking for 6.0.

[rauhul]

I was looking at IdentifiableSet recently, is there a reason to duplicate the API surface of SetAlgebra instead of using Set<Identified<Element>>?

struct Identified<Element> where Element: Identifiable {
  var element: Element
}

extension Identified: Equatable {
  static func ==(lhs: Self, rhs: Self) -> Bool { lhs.id == rhs.id }
}

extension Identified: Hashable {
  func hash(into hasher: inout Hasher) -> Bool { hasher.combine(self.id) }
}

[MaxDesiatov]

How would one look up an element by its ID in Set<Identified<Element>>? We also have subscripts defined on IdentifiableSet, it's a somewhat weird Dictionary/Set hybrid.

[rauhul]

How would one look up an element by its ID in Set<Identified<Element>>? We also have subscripts defined on IdentifiableSet, it's a somewhat weird Dictionary/Set hybrid.

Clearly I didn't think this through enough!

[MaxDesiatov]

For a bit more context, we were storing some of its elements in an Array previously, then for deduplication Set made more sense, and we were also storing some of those elements in dictionaries too. To make those element types Sendable and modifications local and predictable, we had to convert them from classes to structs. The synthesized Hashable and Equatable conformances on structs led to performance regressions. Use of Identifiable and IdentifiableSet were introduced to fix that. Then the only remaining hole was non-determinism when iterating through elements, which is now fixed with OrderedDictionary.