feat: accept the ndt7 server timestamp on measurements

[Victor1Ja]

Stores M-Lab's own clock alongside the device clock on each measurement.
Client side: unicef/project-connect-daily-check-app#85.

Why

• Every measurement is stamped with Date.now() in the app, so measurements.timestamp is only as trustworthy as the clock of the machine running the test — and on these devices it is frequently off by hours or days.
• From app v2.0.4 the ndt7 client also reports M-Lab's clock, read from the Date header of the locate service response.
• This is a second, independent reference point next to timestamp, not a replacement. Nothing that reads timestamp today changes behaviour.

What changed

• Schema: server_timestamp DateTime? @db.Timestamptz(6) on measurements. Not on measurements_failed, matching how scheduled_at / scheduled_slot were handled.
• Migration: 20260824130000_add_server_timestamp_to_measurements — one nullable ADD COLUMN, no default. Additive only; no existing migration touched.
• DTO: optional server_timestamp?: Date with an @ApiProperty description.
• Service: mapped in toModel() and in the DTO mapping (?? null), same pattern as scheduled_at.
• Mocks: field added to the three measurement mocks.

Compatibility

• Optional throughout — every release before v2.0.4 keeps working unchanged.
• Sits after refactor: rename upload_failed to offline_synced and name the slot windows #354's rename migration and touches a different column, so the two are independent.

Caveat worth knowing

The value is M-Lab's clock at server discovery, a few seconds before the test starts, at 1-second resolution. Good enough to detect a device clock that is badly off; not a precise start-of-test timestamp. The ndt7 protocol carries no wall clock of its own — see the client PR for why this is the only reachable source.

Testing

• npx jest — 446 passed, 63 suites (444 on staging before this), no regressions.
• 2 new tests: value persisted when sent, stored as null when omitted.
• npx tsc --noEmit and npx prisma generate clean.