Give the retention sweep tests a larger wait budget on slow CI runners

[ramonsmits]

RetentionSweepTests.A_sweep_cancelled_between_batches_is_recorded_as_cancelled fails now and then on the Linux SQL Server job. The test waits 10 seconds for the manual sweep to stop. In the failing runs the sweep's first batch delete alone took over 10 seconds, so the wait expired while the sweep was still running.

• WaitForManualSweepToFinish now waits up to 60 seconds. The wait returns as soon as the sweep stops, so the larger budget only costs time when the sweep is stuck.
• Same approach as Make the RavenDB index wait in the persistence tests tolerant of slow CI runners #5951, which raised the RavenDB index wait for the same reason.

The slow delete is most likely contention on the shared full-text catalog: every parallel test schema puts its FailedMessages full-text index in the same catalog, and a delete on that table waits for other tests' index DDL. That is a separate change.

[rbev]

I agree that it's probably contention on full text that causes this, and why it's hard to see when you have speedy SSD disks locally.