Method/accessor source text and object-literal accessor names: String(C.prototype.m) is [native code]; get-accessor .name is "" #9468

Follow-ups to #9413 (PR #9465), which added the class-side source registry (Module::class_source_text). Three smaller identity gaps remain, grouped because they share the fix surface:

1. String(C.prototype.m) returns function () { [native code] }; node returns the method source (m() { return 1; }). Class methods compile to perry_method_* symbols rather than closures with a registered source, so the runtime: Function.prototype.toString returns [object Object] for functions (no source reconstruction; no this brand check) #4101 closure source registry never sees them. Needs the method-side equivalent of what PR fix: static this is the constructor, not an instance (#9404); class .name/toString/inspect report source identity (#9413) #9465 built for classes — ast::ClassMethod spans sliced the same way. (Object-literal method toString already works — the PR's objmethod-toString control proves the registry is live for closures.)

2. Object-literal accessor .name is "" where node gives "get g" / "set s". Class accessors are correct ("get p"), so the NamedEvaluation prefixing exists — the object-literal lowering path doesn't call it.

3. .cts NamedEvaluation gaps, upstream of class metadata entirely: module.exports = class {} gets .name === "__perry_cjs_default__" from the pre-parse source rewrite (cjs_wrap/hoist_classes.rs:178; node: ""), and const C = class {} inside a .cts loses NamedEvaluation ("" vs "C"). The committed test_class_name_cjs_9413.cts documents both and deliberately does not assert them — un-skip those arms as part of this fix.

Verification bar

Extend the two #9413 fixtures rather than adding new ones: method/getter/setter/static-method source text through String(), .toString() and template interpolation; async and generator methods; object-literal get/set names; the two .cts arms un-skipped. All byte-compared to node, each demonstrated failing before the change.