Add custom retention procedure

[s0me0ne-unkn0wn]

Partially fixes #29.

Add a custom retention procedure devoid of drawbacks of the stock one:

• It doesn't rollback the entire retention job if the time budget expires, and commits chunk-by-chunk;
• It accounts for its own time budget, not relying on the scheduler to kill the worker, and thus avoids triggering, during normal function, the backoff logic that indeed makes things worse;

The migration also removes the compression policy we don't use, which contributed to the DB stall during the event storm.