Reset the today-just-in fixture through the app, not behind its back

The mobile project read the day as final on a table with no sleep in it.
The rows were gone; what was left was the dashboard snapshot the desktop
run built from them. It carries the sleep-freshness the digest reads, it
lives in the analytics SWR bucket at sixty seconds fresh and an hour of
stale-serve, and a DELETE straight into Postgres does not touch it. So
the second project asked whether last night's sleep was in, and a cached
answer from the first project said yes.

The purge now goes through the bulk-delete route. That tombstones the
rows the way the management list does, recomputes the rollup buckets the
snapshot reads through, and hard-evicts the bucket, so the next read
rebuilds from the table rather than from the run before it. The
delete-by-external-id route names the rows the way this spec does, but it
invalidates without evicting, and a marked-stale cell is served while it
revalidates - the next read would still have been the stale one. The ids
come from SQL because the measurements list collapses overlapping sources
to one row per day and can hide a row that still counts.

The purge stays where it is rather than moving into the cleanup block:
the eviction only happens when the delete removes something, so tidying
up at the end of each run would leave the next one with nothing to delete
and the same stale snapshot.

The starting state is asserted, not assumed, and it stays that way.

e2e/today-just-in.spec.ts

@@ -100,16 +100,58 @@ test("a fresh arrival flips the day in place", async ({ page }) => {
       "UPDATE users SET morning_digest_refreshed_on = NULL WHERE id = $1",
       [userId],
     );
-    // And no sleep sample from an earlier run of THIS spec, which would
-    // land the day on `final` before the scenario starts and make the
-    // provisional assertion below unreachable. Scoped to the external ids
-    // this file writes: it is the only e2e spec that records sleep for the
-    // fixture account (the two v1427 specs mock the route in the browser
-    // and write nothing), so nothing else can be depending on these rows.
-    await client.query(
-      "DELETE FROM measurements WHERE user_id = $1 AND external_id LIKE 'e2e-just-in-%'",
+    // And no sleep sample from an earlier run of THIS spec — the desktop
+    // project runs first and leaves one behind, and the day reads `final`
+    // for as long as last night's sleep is in the record, which would make
+    // the provisional assertion below unreachable on every run after the
+    // first. Scoped to the external ids this file writes: it is the only
+    // e2e spec that records sleep for the fixture account (the two v1427
+    // specs mock the route in the browser and write nothing), so nothing
+    // else can be depending on these rows.
+    //
+    // Deleted through the APP rather than with a `DELETE FROM measurements`,
+    // and that difference is the whole reason this scenario used to fail on
+    // the second project: a fixture that mutates rows behind the app's back
+    // inherits the app's caching. `sleepLastSeenDaysAgo` is read off the
+    // dashboard snapshot, which is an SWR cell in the analytics bucket
+    // (60 s fresh, one HOUR of stale-serve), so the desktop run's
+    // "sleep is in" snapshot outlived the rows it was built from and the
+    // mobile run read `final` from cache against an empty table.
+    //
+    // `bulk-delete` specifically: it hard-evicts that bucket
+    // (`invalidateUserMeasurements(userId, { evict: true })`, pinned by its
+    // own route test), tombstones the rows the way the management list
+    // does, and recomputes the rollup buckets the snapshot reads through.
+    // The delete-by-external-id route would have been the closer fit for
+    // how this spec names its rows, but it invalidates WITHOUT `evict` —
+    // a marked-stale cell is served immediately while it revalidates, so
+    // the next read would still have been the stale `final` one.
+    //
+    // Ids come from SQL rather than from `GET /api/measurements`: that
+    // list collapses overlapping sources to one canonical row per day, so
+    // it can hide a row that is still in the table and still counts.
+    //
+    // Deliberately NOT moved into the `finally` block. The eviction only
+    // happens when the delete actually removes something, so cleaning up
+    // at the end of each run would leave the next one with nothing to
+    // delete, no eviction, and the same stale snapshot.
+    const stale = await client.query<{ id: string }>(
+      `SELECT id FROM measurements
+        WHERE user_id = $1
+          AND external_id LIKE 'e2e-just-in-%'
+          AND deleted_at IS NULL
+        LIMIT 200`,
       [userId],
     );
+    if (stale.rows.length > 0) {
+      const purge = await page.request.post("/api/measurements/bulk-delete", {
+        data: { ids: stale.rows.map((row) => row.id) },
+      });
+      expect(
+        purge.ok(),
+        `purging ${stale.rows.length} stale sleep rows must be accepted (got ${purge.status()})`,
+      ).toBe(true);
+    }
 
     // The account needs a reason to paint a hero that is not the arrival
     // itself. A connection needing a reconnect is that reason: the digest