Sema: Mostly remove one-way constraints

[slavapestov]

I had to revert this last time (#79105) because it broke observable behavior in some cases involving capture list expressions. For now, there doesn't appear to be an easy way to simulate the current behavior of capture lists without one-way constraints.

However, since one-way constraints are no longer needed for performance, we can still rip out the special handling they received in the connected components algorithm. This still changes behavior in a diagnostics test, but the existing expression there appears to be already invalid. The test case from rdar://problem/143338891 still passes:

func rdar143338891() {
  func takesAny(_: Any?) {}

  class Test {
    func test() {
      _ = { [weak self] in takesAny(self) }
    }
  }
}

[slavapestov]

@swift-ci Please smoke test

[slavapestov]

@swift-ci Please test source compatibility

[xedin]

LGTM! I like this narrower change because it means no changes to inference behavior for the left-hand side type variable.

[slavapestov]

@swift-ci Please smoke test macOS