TASK: Allow finer control of accepted read and write latency of FileBackends

[fcool]

This change raises the default maximal wait time for FileBackend operations to 10 ms, while not affecting the performance on filesystems, which are able to react in the previous latency window.

Additionally it minimizes the random factor of waiting, but maintains the randomized acquiring strategy to let eventual races settle, by repeating the randomized wait time until a fixed amount of time has passed.

The amount of time passing is configurable for read and write latency.