fix(metadata-protocol): judge a package publish against the batch's own pending declarations

`publishPackageDrafts` built its author-time validation context from
`engine.registry` alone -- the ALREADY-LIVE universe. A draft is not in that
registry (the write-through runs on `mode: 'publish'`), and the batch's own
promotions do not put it there either: `applyRegistryWriteThrough` lives in
Phase 2, after the Phase-1 transaction in which every draft is gated and
promoted. So while a batch was judged, no member of it was visible to any other
member, in any order.

Measured 2026-08-21 on a cloud rig: a package carrying `dataset/shyx_customer_ds`
and a `dashboard` whose widget binds it rolled back at the dashboard with
`[widget-dataset-unknown] ... does not resolve to a declared dataset`, on every
attempt and under both dataset names the author tried. So a package shipping a
dashboard together with its dataset could never publish.

The batch door now reads its own pending drafts ONCE, before any promotion (a
promote deletes the row it reads), and folds them into all four collections the
closure carries -- `objects`, `permissions`, `books`, `datasets` -- replacing a
live declaration of the same name rather than sitting beside it. A name in
neither the batch nor the live universe is still refused exactly as before.

Fixes #10377

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Author: os-zhuang

.changeset/publish-batch-closure-carries-pending-drafts.md

@@ -0,0 +1,24 @@
+---
+'@objectstack/metadata-protocol': patch
+---
+
+fix: a package publishes as a self-consistent unit — `publishPackageDrafts` judges each draft against the batch's own pending declarations
+
+The batch publish door built the author-time validation context from
+`engine.registry` alone, i.e. the ALREADY-LIVE universe. A draft is not in that
+registry, and the batch's own promotions do not put it there either: the
+registry write-through runs in Phase 2, after the Phase-1 transaction that gates
+and promotes every draft. So while a batch was being judged, no member of it was
+visible to any other member — in any order.
+
+Measured consequence: a package shipping `dataset/x` together with a `dashboard`
+whose widget binds `x` could NEVER publish. `validateWidgetBindings` raises
+`widget-dataset-unknown` at `severity: 'error'`, which refuses the promotion,
+and the batch being all-or-nothing rolls the whole package back. Renaming the
+dataset could not help, and neither could re-ordering the items.
+
+`publishPackageDrafts` now reads its own pending drafts once, before any
+promotion, and folds them into all four context collections the closure carries
+(`objects`, `permissions`, `books`, `datasets`) — pending declarations replace a
+live one of the same name, never sit beside it. A binding that resolves to
+neither the batch nor the live universe is still refused exactly as before.