regression on main: renamed class exports via namespace all resolve to the same class (test_gap_renamed_class_export_namespace) — breaks effect Schema shape

[proggeramlug]

Summary

test-files/test_gap_renamed_class_export_namespace.ts (the #1758/#1812 regression guard) fails on current main — pre-existing, reproduced on pristine main locally and surfacing as conformance-smoke (4) failures on every open PR (e.g. #6545 / #6547), since the smoke suite only runs on PR merges.

expected (node):                          got (perry main):
(1) M.Number.ast._tag: NumberKeyword     (1) M.Number.ast._tag: DirectKeyword
(2) M.Widget.ast._tag: WidgetKeyword     (2) M.Widget.ast._tag: DirectKeyword

Every renamed class export (export { Number$ as Number }, export { Widget$ as Widget }) read via import * as M now resolves to the same class — the direct (non-renamed) export's inherited static ast (DirectKeyword) — instead of each alias's own origin class. The direct-export case (3) still works.

This is the effect-Schema-blocking shape from #1785 (S.Number = class Number$ extends make(...) re-exported as Number; decodeUnknownSync(S.Number) reads S.Number.ast), so the regression likely re-breaks effect Schema decode.

Also failing on pristine main in the same shard: test_gap_zlib_3285_params (zlib params error-code mismatch) — possibly a separate regression from the same merge window.

Suspect window

Both tests passed before the recent merges; candidates are the class-machinery changes in the #6537 (capture-class family) / #6529 window — the per-alias origin-name fallback (class_ids origin recording in compile.rs + namespace-member resolution in property_get.rs, per the test header) may now be preempted by a newer name-keyed path that resolves all aliases to one class.

Verified: fails identically on pristine main (137cfefca-era and 520894f27) with a clean object cache; unrelated to the open class PRs (#6545/#6547) — fails without them.