Sync the 2.4.9 line with upstream develop (bc7899a)

[jeanmarcos-dev]

Merges magento/inventory@develop (bc7899a, 2026-07-30) into the 2.4.9 line, bringing the
distribution back in sync with upstream and turning the next sync into a plain
git merge upstream/develop instead of another manual audit.

The merge-base was b3fba80 (PR magento#703, 2026-04-28), 176 commits behind. Most of that range is
MFTF coverage (ACQE-*) and Adobe's own PR Quality Gates tooling; the functional delta is five
tickets.

Upstream fixes this brings in

• ACP2E-4349 — drops the pinned queue="…" attribute from the <publisher> entries in
  InventoryCatalog, InventoryIndexer and InventorySales. This reverts AC-14558
  (RabbitMQ → ActiveMQ migration). The attribute is read only by module-stomp
  (Model/Publisher.php, Model/MassPublisher.php, both $publisher->getQueue() ?: $topicName),
  so it is inert for AMQP and the MySQL queue. For the seven MSI topics the destination does not
  change either, since every consumer declares a queue named after its topic; the only effective
  change is on the three async.V1.inventory.bulk-product-source-*.POST topics under ActiveMQ.
• ACP2E-5003 — GetStockIdForCurrentWebsite::execute() no longer propagates
  NoSuchEntityException when ?store= carries an unknown code, falling back to the current
  store. That path is hit by the category listing stock plugins, the GraphQL stock-status and
  only-x-left resolvers, and cart merging.
• ACP2E-4741 — InventoryGroupedProductIndexer/Indexer/SiblingProductsProvider now filters
  catalog_product_link by link_type_id = LINK_TYPE_GROUPED, so related, up-sell and
  cross-sell links stop being treated as grouped children. The grouped SelectBuilder already
  applied that filter; only the sibling provider was missing it.
• ACP2E-4765 — the pickup location address now resolves its region by region_id when one is
  set, instead of loadByName, which broke after a Directory subdivision rename.
• ACP2E-4973 — clears redundant in-store pickup data when the shipping carrier changes,
  scoped to the GraphQL area.

ACP2E-4805 and ACP2E-4866 were already cherry-picked here and their files are byte-identical
to upstream, so the merge is a no-op for them.

Conflict resolution

Seven files conflicted:

File	Resolution
InventoryBundleProductIndexer/Indexer/SelectBuilder.php	Trailing whitespace only; kept ours, ->order('sku ASC') untouched
InventoryConfigurableProductIndexer/Indexer/SelectBuilder.php	Kept $enabledChildIsSalable over upstream's MAX(stock.is_salable), kept $eavConfig, and removed a $manageStock block the auto-merge had duplicated
CompositeProductReindexOnNonDefaultStockTest.php, IsConfigurableProductSalableOnNonDefaultStockTest.php	Both add/add. Our copies were byte-identical to the upstream blobs of the commits they were picked from, so upstream's newer versions (real secondary website via ScopeFixture/WebsiteFixture) win
InventoryConfigurableProductIndexer/composer.json, InventoryGroupedProductIndexer/composer.json	Upstream only added inventory-configuration-api, already present and pinned here; kept our pinned block
InventoryInStorePickupQuoteGraphQl/composer.json	Added module-checkout: 100.4.* and module-inventory-in-store-pickup-shipping-api: 1.1.* with this distribution's pinned constraints

The CartManagementPlugin half of ACP2E-4741 was deliberately not applied: it patches
aroundPlaceOrder to fall back from cartRepository->getActive() to get(), and this fork
rewrote that plugin to aroundSubmit(QuoteManagement, callable, Quote, $orderData), which
receives the quote directly and never queries the repository.

CI change

The merge brings .github/scripts/check_copyright_pr.php, a CLI script that legitimately uses
echo/exit and therefore fails the Magento2 standard. The QA gate now excludes .github/ from
its file scan, in both dist-qa.yml and dist-qa-local.sh — the gate covers distributed module
code, not CI tooling inherited from upstream.

Upstream's pr-quality-gates.yml is kept as-is. It is guarded by
if: vars.PR_QA_ENABLED_JOBS != '', which is unset here, so it stays skipped; deleting it would
create permanent divergence and reintroduce a conflict on every future sync.

Verification

Nothing owned by this fork changed. A per-file check over the 314 paths touched only by the fork
reports 0 altered, 0 lost and 0 resurrected; module packaging is unchanged at 74 modules and 148
replace entries.

Run against a real Magento 2.4.9 install with this branch deployed:

• setup:di:compile — clean
• Source-level reservations e2e — 25/25 checks pass (per-source split, cross-stock, shipment,
  cancellation, credit memo, partial shipment, flag-off regression, mid-flight toggle)
• Cross-stock concurrency harness — 20 rounds, exactly one successful order per round,
  0 oversells, 0 deadlocks
• Reservation-path concurrency — 8 workers, 400 place/compensate operations, 0 errors, and every
  (order, source) group balanced
• QA gate (php-lint, phpcs Magento2, phpmd) — pass
• Unit suite — 377 tests, the single failure (GetChildrenSkusOfParentSkusTest::testExecute)
  reproduces identically on dist-2.4.9 without this merge and is caused by array keys preserved
  by array_unique

Three in-store-pickup and source-deduction unit tests abort under PHPUnit 12.5.31 when
createPartialMock is applied to interfaces carrying extension attributes. The same abort
reproduces on the pre-merge tree, and the test files are byte-identical to upstream.

The perf-toolkit leg of the stress suite (concurrent checkouts over 300 multi-source SKUs) was not
run: that dataset does not exist in the 2.4.9 database. Real concurrent checkout behaviour is
covered by the cross-stock harness above; what is missing is scale.