Reclaim Disk Space doc needed

[nerophon]

We need a document that explains, given a backend, how to reclaim disk space both in normal circumstances and in an emergency.

For example, the leveldb explanation is:

In leveldb, there is no GC except compaction. However, compactions cannot be hurried. There are two options to flush old data faster:

full leveldb repair on all partitions (node is offline)
full rolling replace of all nodes (node is online)

We recommend not trying either of these operations while doing other things, for example an upgrade.