Plaid v2 Phase 1: deterministic bundle parent QLI emission

Fix duplicate-row tie-breaking in the Plaid writeback so FCRA bundle umbrellas survive deduplication, and source the parent QLI's list price from the parent SF row instead of the child's override.

1. Why this exists

DEA-6423 reported missing bundle parent QLIs for Plaid's FCRA services bundles and base report when writing back from Dealops v2 to Salesforce CPQ. The umbrella line — the row that stitches child QLIs together via SBQQ__RequiredBy__c — was silently absent on some writes and present on others against the same quote.

2. The duplicate-row scenario

Concrete shape of what the database returns for one FCRA child Product2Id. Both rows match the requested currency; one has a parent link, one doesn't.

Before the fix, isPlaidSalesforceProductRowBetter ranked both rows to the same tuple (no expected parent context, no tiers, no pricebook match), so the comparator returned false and the reducer kept whichever row Postgres emitted last. After the fix, the new hasAnyParent slot breaks the tie in favor of Candidate A.

3. The rank tuple, before / after

The ranking lives in isPlaidSalesforceProductRowBetter. Higher tuple wins, lexicographically. The shape depends on whether the product is tiered.

Tiered products

Non-tiered products

The new hasAnyParent slot is always last, so it only matters when every earlier slot ties. That preserves existing behavior: a row with more TIER children still beats a parent-linked sibling with fewer tiers, because tierChildCount sits at index 1 (tiered) or index 2 (non-tiered) and breaks the tie before hasAnyParent is consulted.

Finally, when even hasAnyParent ties, the comparator falls through to a stable candidate.id < incumbent.id string compare on the Prisma row id — making selection deterministic across runs.

if (candidate.id && incumbent.id && candidate.id !== incumbent.id) {
  return candidate.id < incumbent.id;
}
return false;

4. Parent list price: child override → parent SF row

The second change is independent of the dedup fix. Even when the umbrella QLI was being emitted before, its SBQQ__ListPrice__c and SBQQ__OriginalPrice__c were being sourced from the child's max-ramp override. For FCRA bundles the parent is a $0 umbrella while the child carries the actual $80k service fee — so the parent QLI ended up with the child's price stamped on it.

SBQQ__ListPrice__c:
  listPriceOverride,
SBQQ__OriginalPrice__c:
  parentSfRow?.UnitPrice ?? 0,

SBQQ__ListPrice__c:
  parentSfRow?.UnitPrice
    ?? fallbackListPriceOverride,
SBQQ__OriginalPrice__c:
  parentSfRow?.UnitPrice ?? 0,

The renamed local — listPriceOverride → fallbackListPriceOverride — signals its demoted role: it's no longer the primary source, only the v1-compatibility fallback used when the parent row is unavailable.

5. Test coverage

The bulk of the diff (+297 LOC) lives in PlaidWriteBack.test.ts. Four new tests pin the duplicate-row reducer behavior, plus one new test in parentBody.test.ts and one in the composite-request suite for the list-price change.

6. What this PR doesn't change

7. Risks & open questions