fix(codegen): default-export-of-named-fn wrapper forwards to body (#967)

[proggeramlug]

Summary

function add(a,b){…}; export default add; is the canonical npm-package
barrel idiom (ramda, date-fns, lodash-es, hundreds more). When that
shape was compiled as a native perry.compilePackages module, the
consumer's const fn = add; fn(2,3) returned undefined instead of
5. Direct add(2,3) worked — the bug was specific to closure-form
access (alias to local, callback passing, indirect call).

Root cause: codegen's sub-bug-B wrapper-alias loop emitted a NO-OP
__perry_wrap_perry_fn_<src>__default wrapper for every
Export::Named { local, exported } where local == exported and
local wasn't a HIR function name. The branch's original purpose was
to make import * as z; export { z }; (re-exported namespace) and
export { SomeClass } link by emitting an → undefined thunk. But the
default-export shape ALSO satisfies local == exported == "default"
with func_by_local_name.contains_key("default") == false — because
the HIR function's name is add, not default. So the no-op fired
and short-circuited every closure-form call through the default
import.

The HIR lowerer at perry-hir/src/lower.rs:5601 correctly registers
the alias in module.exported_functions.push(("default", add_id)).
The fix consults that map: when a local==exported no-function-by-
local-name export resolves to an aliased real function, emit a
forwarding wrapper (closure_ptr → tail-call perry_fn_<src>__add)
instead of the no-op. Mirrors the local != exported branch at L2792
which already handles renamed exports correctly.

Test plan

• New test-files/test_default_import_as_value.ts +
  _helper.ts cover four value-position shapes: local alias,
  function-parameter pass-through, closure-capture, direct call.
  Output matches node --experimental-strip-types byte-for-byte
  (5 / 30 / 15 / 5).
• test-files/test_lodash_default_import_methods.ts (issue feat(security): #498 — pin SHA-256 of perry.nativeLibrary prebuilt archives in perry.lock #957
  regression) still passes.
• test-files/test_issue_678_reexport_default.ts still passes.
• CI: lint / cargo-test / parity / compile-smoke /
  api-docs-drift / security-audit once green.

Out of scope (related but separate bugs)

With this fix applied, two packages from the original compat sweep
still don't fully work:

• ramda R.sum([1..5]) returns [object Object] because
  Function.prototype.apply has no runtime handler — fn.apply(this, args) returns the receiver as a stub instead of invoking fn.
  Triggers inside ramda's _curry1/_curry2/_curry3 machinery.
  Untriaged.
• date-fns format(new Date(), 'yyyy-MM-dd') throws
  RangeError: Invalid time value because Date.prototype.constructor
  returns undefined, breaking constructFrom's
  new date.constructor(value) cloning. Untriaged.

Both packages exercise the default-import-as-value shape internally
many times, so this fix is a prerequisite for any future work on
them — without it the closure-dispatch would short-circuit before
reaching the apply / constructor failures.

Version

0.5.980 → 0.5.981