fix: nargo expand bugs behind ignored integration tests

[asterite]

I figured that nargo expand is a good tool to understand what the actual code is, so I thought that maybe it's a good idea if we have this working well in (almost) all cases. This PR fixes the remaining bugs we had.

Description

Problem

Resolves the nargo expand bugs tracked by the ignore lists in tooling/nargo_cli/build.rs: IGNORED_NARGO_EXPAND_COMPILE_SUCCESS_EMPTY_TESTS, IGNORED_NARGO_EXPAND_COMPILE_SUCCESS_NO_BUG_TESTS and IGNORED_NARGO_EXPAND_EXECUTION_TESTS — tests that were ignored because the expanded code failed to recompile (or recomputed the wrong value).

After this PR, every remaining ignored entry is a deliberate test-infrastructure limitation (no src/main.nr, workspace layouts, relative-path or transitive-only dependencies), not a bug.

Summary

One commit per fix:

1. Print Self::item for assumed trait items. Inside a trait's default method body, Self::static_method_2() printed as ATrait::static_method_2(), which doesn't resolve on recompile (unknown receiver type). The printer now tracks the trait's Self type variable while printing its body. Un-ignores trait_function_calls and trait_static_methods.

2. Don't print impl Trait parameter constraints in where clauses. An impl Trait parameter desugars to a hidden generic named impl {Trait} plus a trait constraint, which printed as where impl SomeTrait: SomeTrait. The constraint is implied by the parameter type, so it's now filtered out. Un-ignores trait_method_mut_self.

3. Print impl Trait parameters with their trait generic arguments. The hidden generic's synthetic name omits the trait's generics, so _input: impl Foo<N> printed as _input: impl Foo. Such parameters are now printed from their resolved desugared constraint. Un-ignores regression_7648.

4. Don't embed "(resolved type)" in associated type names. The synthetic name of an associated type's named generic (<{object} as {trait}>::{name}) was built by stringifying the object type as written; a macro-spliced UnresolvedTypeData::Resolved displays as the (resolved type) placeholder, producing unparseable output like <(resolved type) as Ser>::N. These names now look through the resolution to the actual type. Un-ignores regression_10747_associated_constant and (incidentally) the regression_9116 execution test.

5. Print a global's initializer when its value can't be printed as code. Globals printed their comptime-evaluated value even when that value doesn't reconstruct as compilable source: struct literals whose type or fields are private outside their defining module (BoundedVec { len, storage }, Option { _is_some, _value }), comptime-only values that print as a panic(...) placeholder, or (commit 12) strings whose bytes aren't valid UTF-8 and would print lossily with a different length. For those, the original initializer expression is printed instead; representable values keep printing as values, since a comptime-mutable global's final value can differ from its initializer. Un-ignores trait_call_in_global, function_registry, regression_10887 and regression_12269.

6. Qualify associated constant references. An expression reference to a trait impl's associated constant (Self::N) printed as the bare name N, which doesn't resolve on recompile. It now prints as Self::N inside the defining impl and <Type as Trait>::N elsewhere; when the trait isn't even visible from the printing module (a comptime Expr::resolve in another module's scope can splice such a reference), the constant's compile-time value is printed instead. Un-ignores trait_associated_constant, regression_10466 and comptime_resolve_associated_constant_scope.

7. Fix numeric type alias printing. Numeric aliases printed without their numeric type annotation (type Double<let N: u32> = N * 2; instead of ...: u32 = N * 2;), and an alias parameter used as a value (AliasN::<1>) printed as the bare name N — which silently resolves to an unrelated item with the same name, changing the program's meaning (this one is a red/green commit pair: the buggy snapshot is recorded first). Un-ignores numeric_type_alias.

8. Print associated constant declarations that involve the self type. let N: i32 = -12345i32; in impl Foo for i32 printed as let N: Self = -12345; — Self doesn't resolve in that annotation, and the unsuffixed literal is checked as u32. Both the annotation and the value's type suffix are now printed. Un-ignores negative_associated_constants.

9. Cleanup: 15 noirc_frontend_tests_* entries in the compile_success_no_bug ignore list referred to test programs deleted in chore: remove duplicated frontend tests #9706 and ignored nothing; removed. The ignore lists' doc comments now reflect that only deliberate limitations remain.

Each fix comes with a round-trip regression test in compiler/noirc_frontend/src/tests/expand.rs where the harness allows it (a features-aware assert_no_errors_and_to_string_using_features variant was added for the impl Trait tests), plus the un-ignored integration tests.

Additional Context

Verified with the full nargo_cli execute suite (including all 789 nargo_expand tests and the compile-failure error-message snapshots, relevant since commits 4 and 6 touch elaborator-minted names), the noirc_frontend suite, and clippy.

🤖 Generated with Claude Code

Review follow-ups

Two additional commits address the review findings:

• fix: suffix instantiated numeric generics with their numeric type — the printed constant for an instantiated numeric generic was unsuffixed, so an inference site could silently select a different trait impl (wrong program that still compiles). Adds the numeric_type_alias_inference execution test that fails at runtime without the suffix.
• fix: harden nargo expand's unrepresentable-value detection — strings must survive {:?} formatting (control chars/combining marks escape as \u{..}, which Noir does not lex), format-string fragments are checked for raw-printed specials, capturing closures are unrepresentable, plain function values get the same visibility check as structs/enums, a warning comment marks a mutable global whose printed initializer cannot reproduce its final mutated value, the impl {Trait} synthetic-name prefix is now a shared constant, and the Display limitations of the minted associated-type names are documented. Adds the comptime_mutated_global and global_string_control_char execution tests.

[TomAFrench]

Claudebox found a few issues on this PR: https://gist.github.com/AztecBot/018be6562f9cb4ecbf81a76bd14ac8f2