Flaky test: tests/snuba/tasks/test_unmerge.py::UnmergeTestCase::test_unmerge

Failing test: tests/snuba/tasks/test_unmerge.py::UnmergeTestCase::test_unmerge
Sentry sha: 405c781e4bcb9380607169cc7de045bcc30823e6
Run: https://github.com/getsentry/sentry/actions/runs/22270628212

tests/snuba/tasks/test_unmerge.py:301: in test_unmerge
    assert list(
E   assert [(7, datetime...imezone.utc))] == [(6, datetime...imezone.utc))]
E     
E     At index 0 diff: (7, datetime.datetime(2026, 2, 22, 4, 47, 37, tzinfo=datetime.timezone.utc), datetime.datetime(2026, 2, 22, 4, 47, 43, tzinfo=datetime.timezone.utc)) != (6, datetime.datetime(2026, 2, 22, 4, 47, 37, tzinfo=datetime.timezone.utc), datetime.datetime(2026, 2, 22, 4, 47, 42, tzinfo=datetime.timezone.utc))
E     
E     Full diff:
E       [
E           (
E     -         6,
E     ?         ^
E     +         7,
E     ?         ^
E               datetime.datetime(2026, 2, 22, 4, 47, 37, tzinfo=datetime.timezone.utc),
E     -         datetime.datetime(2026, 2, 22, 4, 47, 42, tzinfo=datetime.timezone.utc),
E     ?                                                ^
E     +         datetime.datetime(2026, 2, 22, 4, 47, 43, tzinfo=datetime.timezone.utc),
E     ?                                                ^
E           ),
E       ]

Start with tests/snuba/tasks/test_unmerge.py, especially test_unmerge around line 301, and reproduce the failure from the linked GitHub Actions run. Trace why the result differs in count and timestamp, then verify the test passes consistently without changing its intended assertions.