fix(runtime): bind this to receiver in js_get_iterator for yield*-over-a-Context-Tag (#321)

[proggeramlug]

Problem

Effect.gen over a Context.Tag yielded empty. The last effect survey blocker:

const program = Effect.gen(function* () {
  const g = yield* Greeter;        // Greeter is a Context.Tag class
  return g.greet("Ada");
});
const result = Effect.runSync(Effect.provide(program, GreeterLive));
console.log("ctx/layer:", result);  // Node: "ctx/layer: Hello, Ada"  Perry: (empty)

Effect.map over a Tag and Layer DI already worked; basic Effect.gen + yield* Effect.succeed(x) already worked. The break was specifically yield* <ContextTag> — a Tag is a class-as-value that is itself a yieldable Effect.

Root cause

A Context.Tag is yieldable because Context.Tag(id) sets the class's prototype to TagProto = { ...EffectPrototype, ... }, and EffectPrototype[Symbol.iterator] is:

[Symbol.iterator]() { return new SingleShotGen(new YieldWrap(this)); }

This is an object-literal method inherited via the prototype chain whose body reads this. Perry codegen bakes this for object-literal methods to the defining literal (the CAPTURES_THIS_FLAG reserved slot — correct for [Symbol.toPrimitive] reading own fields). When the yield* desugar calls js_get_iterator(Greeter), the inherited [Symbol.iterator] was invoked via js_closure_call0 with no receiver, so this stayed the prototype object instead of the Tag.

Consequence: new YieldWrap(this) wrapped the prototype, not the Tag. yieldWrapGet(state.value) in effect's fiber OP_ITERATOR loop returned a non-Effect (_op was undefined), so the fiber never resolved it to the service and never resumed the generator — empty output.

Isolated with a manual generator drive (no fiber):

• Node: yieldWrapGet(it.next().value)._op === "Tag"
• Perry (before): ..._op === undefined

And reproduced with no effect at all (object-literal prototype method reading this, called with a receiver -> Perry bound this to the prototype, Node to the receiver).

Fix

In js_get_iterator, rebind the resolved [Symbol.iterator] closure's this to the original iterable (via the existing clone_closure_rebind_this) before calling it — matching the spec that iterable[Symbol.iterator]() runs with this === iterable. It is a no-op for iterator methods that do not capture this (real generators, arrays, Map/Set/string iteration unaffected).

One runtime file changed (crates/perry-runtime/src/symbol.rs, +12/-1) plus a standalone gap test.

Verification

• survey/real3_context_layer.ts -> ctx/layer: Hello, Ada (was empty)
• survey/nlay.ts -> provide: 42 (Layer DI, no regression)
• survey/05_gen.ts -> 30 (no generator regression)
• survey/ctx_works.ts -> provide: 42; minimal Effect.gen + Context.make (no Layer) -> min: Hi Ada
• 12 other effect survey items (succeed/map/flatMap/pipe/fail_catch/all/sync/either/option/data/nref/real4_ref) byte-identical to Node
• test-files/test_gap_yieldstar_inherited_iterator_this.ts (standalone, no effect) byte-identical to Node
• Iteration regression sweep (for-of array/gen/Map/Set/string, yield*-over-generator, spread) byte-identical to Node
• cargo test --release -p perry-runtime --lib -- --test-threads=1 -> 690 passed, 0 failed
• cargo fmt --all -- --check clean