runtime: make Set constructor consume arbitrary iterables and throw on non-iterables

[andrewtdiz]

Summary

new Set(iterable) currently routes through js_set_from_iterable, but that helper only handles a few input shapes directly:

• arrays,
• strings,
• null / undefined as empty,
• other pointer-tagged values by treating them as arrays,
• everything else as an empty set with no error.

Node's Set constructor should consume any iterable and should throw TypeError for non-null, non-undefined values that are not iterable.

Node behavior

Checked with Node v25.9.0:

[...new Set([1, 2, 1])];
// [1, 2]

[...new Set(new Set([3, 4]))];
// [3, 4]

[...new Set('aba')];
// ['a', 'b']

[...new Set(new Map([[1, 'a'], [2, 'b']]))].map(v => v.join(':'));
// ['1:a', '2:b']

const customIterable = {
  *[Symbol.iterator]() {
    yield 'x';
    yield 'y';
  },
};
[...new Set(customIterable)];
// ['x', 'y']

new Set(null).size;      // 0
new Set(undefined).size; // 0

new Set(1);
// TypeError: number 1 is not iterable (cannot read property Symbol(Symbol.iterator))

new Set({ a: 1 });
// TypeError: object is not iterable (cannot read property Symbol(Symbol.iterator))

new Set(Symbol('x'));
// TypeError: symbol is not iterable (cannot read property Symbol(Symbol.iterator))

Perry implementation

Lowering/codegen sends any provided Set constructor argument to js_set_from_iterable:

• crates/perry-hir/src/lower/expr_new.rs:304 handles new Set() / new Set(iterable).
• crates/perry-hir/src/lower/expr_new.rs:316 returns Expr::SetNew for no arguments.
• crates/perry-hir/src/lower/expr_new.rs:318 returns Expr::SetNewFromArray(firstArg) for any argument.
• crates/perry-codegen/src/expr/logical_collections.rs:551 routes Expr::SetNewFromArray through js_set_from_iterable.

The runtime helper is intentionally lenient rather than Node-compatible:

• crates/perry-runtime/src/set.rs:786 documents js_set_from_iterable as creating a Set from any iterable JS value.
• crates/perry-runtime/src/set.rs:788 lists the actually handled shapes: Array, String, undefined/null, and anything else.
• crates/perry-runtime/src/set.rs:795 explicitly says anything else becomes an empty set with no error, even though JS would throw TypeError.
• crates/perry-runtime/src/set.rs:859 treats pointer-tagged values as arrays; non-array objects can be misread as ArrayHeader values rather than using their iterator or throwing.
• crates/perry-runtime/src/set.rs:866 returns a fresh empty set for undefined/null/number/etc., so new Set(1) cannot throw the required TypeError through this path.

Expected fix direction

new Set(iterable) should implement the Set constructor's iterable consumption semantics:

• keep null and undefined as empty-set inputs,
• consume arbitrary iterable objects through their iterator,
• preserve existing array and string fast paths when they match iterator behavior,
• support Map/Set/custom iterables by consuming yielded values,
• throw TypeError for non-iterable numbers, symbols, and plain objects.

Duplicate search

Searched existing issues/PRs for:

• Set constructor iterable
• new Set iterable
• Set constructor TypeError
• new Set non-iterable
• js_set_from_iterable TypeError
• new Set Map iterable
• new Set custom iterator
• PR search for Set constructor iterable OR js_set_from_iterable

No existing issue covered this Set constructor behavior. Related but not duplicate: #2770 tracks Map constructor iterable/malformed-entry semantics, and older Map/Set iteration bugs (#302, #311, #412) were about for-of/spread behavior rather than constructor validation and arbitrary iterable init.