fix(semantic): const_values_eq must canonicalize felt252 inside compounds - #10037

Summary

const_values_eq previously only handled the case where both sides were ConstValue::Int, returning false for any other variant pair. This meant that felt252 values nested inside structs, enums, or NonZero wrappers were compared by identity rather than by field-element value, causing -1 and PRIME - 1 to be treated as unequal even though they represent the same field element.

The function now recurses into ConstValue::Struct, ConstValue::Enum, and ConstValue::NonZero variants, applying field-element canonicalization at every level where a felt252 integer may appear.

Type of change

Please check one:

• Bug fix (fixes incorrect behavior)
• New feature
• Performance improvement
• Documentation change with concrete technical impact
• Style, wording, formatting, or typo-only change

Why is this change needed?

The PartialEq operator for felt252 must agree with the field, meaning -1 and 0x800000000000011000000000000000000000000000000000000000000000000 (i.e., PRIME - 1) are the same value. The previous implementation only canonicalized top-level felt252 integers, so equality checks on NonZero<felt252>, tuples containing felt252, or enums wrapping felt252 would incorrectly return false when comparing these two representations.

What was the behavior or documentation before?

Constant expressions such as:

const VALID_EQ_NONZERO_FELT_CROSS_FORM: () = assert(NZ_FELT_NEG_ONE == NZ_FELT_FIELD_NEG_ONE);
const VALID_EQ_TUPLE_FELT_CROSS_FORM: () = assert((-1_felt252, 5) == (0x800...000, 5));
const VALID_EQ_ENUM_FELT_CROSS_FORM: () = assert(Option::Some(-1) == Option::Some(0x800...000));

would fail constant evaluation because the structural wrappers were compared by identity rather than recursively by field value.

What is the behavior or documentation after?

The above constant expressions now evaluate correctly. const_values_eq recurses into structs, enums, and NonZero wrappers, canonicalizing felt252 integers at every nesting level before comparing.

Related issue or discussion (if any)

N/A

Additional context

Test cases for all three new cross-form equality scenarios (NonZero<felt252>, tuple, and enum) have been added to the constant expression test data.