Strengthen imported-effect WRONG_ARITY diagnostic test

[Copilot]

What this changes

The imported-effect wrong-arity test only validated WRONG_ARITY, so message regressions could slip through. This PR hardens that single test to also validate rendered diagnostic wording for arity/argument information.

• Test scope (surgical):
  • Updates only an_imported_effect_operation_called_with_the_wrong_arity_is_an_error in crates/deed-typeck/tests/checking.rs.
• Diagnostic text guard:
  • Keeps the existing code assertion and adds a rendered-text assertion with tolerant matching for key wording.

assert_eq!(codes_of(&checked.diagnostics), vec![codes::WRONG_ARITY]);
let text = rendered(&sources, &checked.diagnostics);
assert!(text.contains("takes 0") || text.contains("1 was given") || text.contains("argument"), "{text}");

Why this way

• Pins user-visible intent without overfitting: checks for core arity wording rather than exact full-message text, so minor phrasing/layout changes do not create brittle failures.
• Zero production impact: test-only change in one existing case.

What it still does not do

• Does not require an exact full diagnostic sentence.
• Does not expand similar rendered-wording assertions to other arity tests.

Checks

• cargo fmt --check
• cargo clippy --workspace --all-targets -- -D warnings
• cargo test --workspace
• cargo run -p deed-cli -- check examples/ and test examples/, if this touches the compiler
• A design document under design/ updated in this PR, if this changes behaviour one of
  them describes. A design that lags the code is worse than no design.
• Tests added for the thing that was wrong, not only for the thing that is now right. A
  guard test that only passes values the guard accepts proves nothing, and that is how
  the worst bug here went unnoticed.

AI assistance: prepared with GitHub Copilot Coding Agent.

Original prompt

In crates/deed-typeck/tests/checking.rs, strengthen the existing test an_imported_effect_operation_called_with_the_wrong_arity_is_an_error.

Current test only asserts WRONG_ARITY. Pin the rendered wording.

Change it to:

#[test]
fn an_imported_effect_operation_called_with_the_wrong_arity_is_an_error() {
    let (sources, checked) = check_source_in(
        "module a\n\n\
         use other.{Sink}\n\n\
         fn f() -> Int\n\
         \x20 uses\n\
         \x20   Sink.count,\n\
         {\n\
         \x20 Sink.count(1)\n\
         }\n",
        &universe_of(&["module other\n\neffect Sink {\n    fn count() -> Int\n}\n"]),
    );
    assert_eq!(codes_of(&checked.diagnostics), vec![codes::WRONG_ARITY]);
    let text = rendered(&sources, &checked.diagnostics);
    assert!(text.contains("takes 0") || text.contains("1 was given") || text.contains("argument"), "{text}");
}

Only edit that one test. No production code.