Completion for tuple struct's indices is problematic

$ rust-analyzer --version
rust-analyzer 1 (825833c26 2023-03-19)
$ rustc --version
rustc 1.68.0 (2c8cc3432 2023-03-06)

Context: https://lists.gnu.org/archive/html/bug-gnu-emacs/2023-03/msg01295.html and João's response.

With this snippet:

// src/lib.rs
/// 0..=10
struct Foo(u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32);

fn foo(f: Foo) -> u32 {
    f.
}

Move cursor after "f." and type, for example, "1", and then issue completion. Apparently, according to João's test and analysis, RA tells the client that "0" and "1" are possible completion candidates for the position after "f.1".

This set of completion candidates is wrong: the possible candidates should be the following:

• "" (because "f.1" is a valid field)
• "0" (because "f.10" is also a valid field)
• definitely not "1" (because "f.11" is not a valid field).

Reproduce the issue from the src/lib.rs snippet by placing the cursor after f.1 and requesting completion. Trace the tuple-field completion entry point and add coverage for the expected candidates: an empty completion, 0 for field 10, and no 1 for nonexistent field 11.