fix: #809 — object literal spread+methods+computed-keys; Object.create wires prototype

[proggeramlug]

Closes #809. Refs #321, #711.

The Effect end-to-end repro (#321) advanced past ParseResult.ts and threw TypeError: value is not a function in HashRing.ts__init. The 20-line standalone repro from the issue (Proto = { [TypeId]:TypeId, [Symbol.iterator]() {}, pipe() {}, ...Inspectable.BaseProto, toJSON() {} } + Object.create(Proto).toJSON()) printed keys: 2 (node: 4 — the issue's "5" predates checking against node, which excludes Symbol.iterator from Object.keys), inst.toJSON() → \"1970-01-01T00:00:00.000Z\", inst[TypeId] → undefined.

Four independent bugs in the chain, all needed to pass

1. Object-literal has_spread lowering dropped entries. crates/perry-hir/src/lower/expr_object.rs's spread path lowered to Expr::ObjectSpread { parts } whose parts only carries static-string KeyValue + spreads; Prop::Method fell into _ => {} and computed KeyValue keys into _ => continue — so [Symbol.iterator](), pipe(), the [TypeId] computed string key, and the trailing toJSON() were all silently discarded.
2. Over-eager DateToJSON lowering. inst.toJSON() where inst's static class is unknown rewrote to DateToJSON(inst); the runtime read the pointer's NaN-box bits as a timestamp → epoch.
3. js_object_create ignored its argument. Allocated a bare empty object, dropping proto entirely. Object.create(P).x saw nothing.
4. Prototype-walk gaps exposed once (3) was fixed. The codegen PIC fast path for obj.prop spuriously "hit" when obj->keys_array == null (an Object.create result) because the zero-init per-site cache matched, returning slot[0] without invoking the miss handler. js_object_get_field_by_name early-returned undefined on null keys_array before its class_id prototype-walk. js_native_call_method's prototype-object walk was gated inside if let Some(ref reg) = *CLASS_VTABLE_REGISTRY — skipped entirely for programs with no user classes.

Fix

1. Rewrote the has_spread path as a source-ordered IIFE ((__o) => { …ordered ops…; return __o })({}) so static props, computed keys, this-binding methods, and spreads interleave correctly (a later prop/spread overrides an earlier same key — the non-spread fast path can't be used because it applies all static props before any post-init, which would let a trailing ...src clobber the literal's own toJSON()). Spreads emit js_object_assign_one; static this-methods a new js_object_set_method_by_name runtime helper (string-key analog of js_object_set_symbol_method); computed this-methods reuse js_object_set_symbol_method; everything else IndexSet. Extracted resolve_keyvalue_key / lower_method_prop shared helpers used by both spread and non-spread paths (non-spread behavior identical).
2. static_receiver_class now returns Some(\"Object\") for object-literal / Object.create(...) receivers and locals in a new LoweringContext.plain_object_locals set (populated in lower_var_decl_with_destructuring); the ambiguous-Date-method gate treats it like Some(\"URL\") and skips the Date arms.
3. js_object_create now allocates a synthetic class_id, maps it to proto in CLASS_PROTOTYPE_OBJECTS (reusing the perry-runtime: Effect framework throws TypeError: pipe is not a function during Schema.ts module init (Effect end-to-end blocker, post-#685) #711 machinery), and stamps the object. Object.create(null) / non-object / builtin-backed sources fall back to the original prototype-less object.
4. PIC hit predicate now requires keys_array != 0 so keyless receivers route to the slow path. js_object_get_field_by_name's null-keys_array arm consults the prototype chain via a new shared resolve_proto_chain_field helper (which also de-duplicates the existing inline perry-runtime: Effect framework throws TypeError: pipe is not a function during Schema.ts module init (Effect end-to-end blocker, post-#685) #711 walk). js_native_call_method runs an independent resolve_proto_chain_field resolution after the registry walk, with IMPLICIT_THIS set so methods see the right receiver.

Validation

• Issue's exact repro byte-identical to node --experimental-strip-types: keys: 4, inst.toJSON: {\"_id\":\"HashRing\",\"x\":1}, inst[TypeId]: ~test/HashRing.
• Gap suite 34/36 (2 = pre-existing console.dir/group + lookbehind regex categorical gaps, unchanged).
• cargo test --release for perry-hir / perry-codegen / perry-runtime all green (0 failed).
• 12-file curated smoke over spread/object/prototype test-files (incl. test_issue_711_function_prototype, test_spread, test_edge_rest_spread_defaults, test_get_prototype_of_instance) byte-identical to node.
• Targeted regression confirms real new Date(0).toJSON()/toISOString() still work, spread override ordering preserved ({a:1,...{a:2}}→{a:2}, {...{a:2},a:1}→{a:1}), Object.create(null), empty-object missing-prop, Object.create(proto).method() with this-binding all match node.

Known remaining pre-existing gaps (not in #809's DoD, not regressions)

• Symbol-keyed property inheritance through an Object.create prototype chain (o[sym] where sym is on the proto) still returns undefined — the symbol side-table is keyed by receiver pointer and the prototype walk only resolves string keys. Not exercised by perry-codegen: object literal with computed-key props + computed-key methods + cross-module spread drops keys & mis-resolves methods — Effect HashRing.ts blocker (post-#740) #809 (Object.keys excludes symbols; inst[TypeId] is a string key) and Object.create returned nothing before this commit, so it never worked there.
• Object.getPrototypeOf(Object.create(p)) === p still returns the class-ref shim, not p.

Both tracked separately.