Operations in a long retry loop sleep indefinitely, so if they need to be interrupted for step-down/shutdown they have to wait for the next iteration of the WCRetry loop, which could be ~hours in practice.

In order to avoid complications of casting sub-ms sleeps to either 0 or 1ms, we only sleep interruptably when above 10ms. This is sufficient for our SLOs, whilst not changing behaviour for sufficiently short sleeps.