[BACK-4528] Skip patient summary updates older than the stored stats

[toddkazakov]

Why

UpdatePatientSummary (POST /v1/patients/{patientId}/summary) unconditionally overwrites the stored cgmStats/bgmStats. Its reporters deliver at least once with no ordering guarantee, which allows a delayed report to arrive after one carrying a newer calculation and regress the stored stats until the user's next calculation.

Today's CDC consumer (clinic-worker patientsummary) is protected by Kafka partition ordering, but the upload-postprocess work processor in platform (which is taking over summary reporting — see platform upload-postprocess-work) is not: a processor reaped mid-report (>5-minute stall) can complete its UpdatePatientSummary after another pod's retry already pushed newer stats. The EHR sync generates reports from this stored copy, so a regression is visible downstream.

What

UpdateSummaryInAllClinics now applies each stats type only when the incoming dates.lastUpdatedDate (the calculation date) is not older than the stored one, using a $cond pipeline update per type — each patient document and each stats type is compared independently, in the same single UpdateMany.

Rules (falling out of BSON comparison order, missing < null < date):

incoming date	stored stats	result
any	absent	applied
newer or equal	dated	applied
older	dated	kept (regression refused)
none	dated	kept
none	undated	applied

Equal dates still apply, so an at-least-once reporter re-delivering the same report remains idempotent. updatedTime is still always bumped and the ErrSummaryNotFound (204) semantics for users who are not a patient of any clinic are unchanged, as is the nil-body unset path. The incoming stats are wrapped in $literal so they are stored verbatim rather than evaluated as aggregation expressions.

Tests

New repository specs (against Mongo): stores when no stats exist; replaces on newer and on equal calculation dates; keeps stored stats against an older report and against an undated report; applies per type independently in one call (cgm replaced while bgm kept); ErrSummaryNotFound for non-patients. go test ./patients/... green.

[ewollesen]

This test doesn't appear to match its name. It's not clear which was intended.

The test's name seems to indicate its testing the case where neither the existing summary or new summary have dates. The test itself however, tests a case where there's an existing date, but the new summary provides no date.

Maybe "keeps stats with a calculation date away from an update without one" is better?