Fix tuple struct pat expected type

[A4-Tacks]

Example

struct Foo(Option<i32>);
fn foo(x: Foo) -> Foo {
   match x { Foo($0) => () }
}

Before this PR

ty: Foo, name: ?

After this PR

ty: Option<i32>, name: ?

[ChayimFriedman2]

I'm really not pleased with the ways the expected type inference in completion is going. We're essentially duplicating the entire logic of type inference. I don't think that's sustainable.

We need to find a better way, and I hesitate to merge more PRs in this area until we have a plan. A possible half-solution is to wait for speculative execution, then query the type of the inserted identifier.

@rust-lang/rust-analyzer thoughts?

[lnicola]

Sounds like a good idea, but I think we can merge this one since A4-Tacks already did the work and it's not very complex.

[A4-Tacks]

then query the type of the inserted identifier.

Is inference on the modified syntax tree beneficial for caching?

[ChayimFriedman2]

No, it is beneficial for completion (only possible when Salsa will gain speculative execution).

[A4-Tacks]

Can this PR be merged now?

[A4-Tacks]

This bug quite affects the user experience. I often type Some(a$0) and backspace

Similar tuple also has this bug, but the frequency of encountering it is far lower than this