fix(runtime): link Object.create synthetic class_id to original class chain (#1805)

[proggeramlug]

Summary

Fix #1805 — Object.create(Object.getPrototypeOf(instance)) produced
an object that wasn't instanceof the original class, even though
property reads and getter dispatch through the synthetic-prototype
chain worked correctly (the #711 / #809 wiring).

js_instanceof's chain walk keys off get_parent_class_id, but the
synthetic class id allocated by js_object_create had no parent edge
linking it to the original class — so the walk stopped at the
unregistered synthetic id and returned false.

Fix: at synthetic-cid registration time, also record (synthetic_cid → proto_ptr.class_id) in CLASS_REGISTRY when the proto pointer
carries a non-zero class_id. In Perry's model
Object.getPrototypeOf(instance) returns the instance itself (see
js_object_get_prototype_of), so proto_ptr here is a real class
instance whose class_id field IS the user class's id. Linking it as
the synthetic cid's parent lets the chain walk reach the original
class; chained Object.create cases keep working because the walk
transitively reaches the root via synthetic_2 → synthetic_1 → ClassId.

Surfaced by effect's SchemaAST.annotations, which clones AST nodes
via Object.create(Object.getPrototypeOf(ast), Object.getOwnPropertyDescriptors(ast)). Refs #1758, #809.

Test plan

• test-files/test_issue_1805_object_create_instanceof.ts covers
  plain Object.create(proto), the descriptor form, the chained case,
  and the Object.create(null) negative — byte-identical with Node.
• cargo fmt --all -- --check clean.
• cargo test --release -p perry-runtime --lib — 2 flaky failures
  (gc::tests::copying::test_copying_minor_rewrites_exact_object_pointer_slot_only,
  object::tests::text_encoding_stream_globals_construct_readable_writable_shape)
  both pass in isolation; pre-existing thread-state flakiness.