Swift: Dataflow for keypaths

[MathiasVP]

This PR implements dataflow-support for key-path expressions. This has been made possible thanks for Sasha's recent PR which extracted the right things 🎉.

After this PR, we model a key-path expression as a lambda with a sequence of read steps. For example, consider this snippet:

func testKeyPath() {
  let s = S(x: source())
  let f = \S.x
  sink(s[keyPath: f])
}

this is now modelled like we model:

func testKeyPath() {
  let s = S(x: source())
  let f = { (s : S) in s.x }
  sink(f(s))
}

There's still some future work to be done: I've only supported simple field lookups inside those key-path chains. We also need to handle wrapping and unwrapping of optionals, and writeable key-path expressions. All of this should be fairly straightforward once the infrastructure from this PR is merged, though.

Commit-by-commit review recommended.

@d10c I've assigned you as a reviewer because I think you've touched many of these files before (and will probably be touching them again soon). Let me know if you'd like to go through this PR over Zoom or something.

[MathiasVP]

DCA looks fine: the new lost result is due to extractor-wobbliness. When I run SideStore/SideStore locally, I see the result when running this PR on the database produced for the main run.

[geoffw0]

Results LGTM but I would like to see more diverse test cases. You mentioned writeable key-path expressions and optional handling as planned follow-up. I've also come across:

• keypaths of the form \.field where the object can be inferred from context
• keypaths into arrays \[Int][0]

[MathiasVP]

Sure, I'll add more tests. I wasn't aware of the inferred-context version. I'm also sure \[Int][0] won't work currently, but will work automatically once we add support for ArrayContent (but obviously I might as well add the tests now).

[MathiasVP]

231b0fc adds a few more tests.

[geoffw0]

Great, thanks. Happy with the tests now!

I'm also sure [Int][0] won't work currently, but will work automatically once we add support for ArrayContent (but obviously I might as well add the tests now).

Good point. It will be nice when we have ArrayContent working and this comes together.

[d10c]

Looking good! See my comment though.

[d10c]

I think you should have all the pieces now for making this work, no? We have content flow for optionals, it just needs to be linked to key paths like you did with field contents.

I'm also fine with leaving this as a follow-up issue though, but if it's not too complicated it would be nice to see this test pass.

[MathiasVP]

Yeah, I think everything should be in place to make this work. I'd like to save it as follow-up, though :)

I'll create an issue with what needs to happen as a follow-up