Compactions improvements

[markpapadakis]

• Allow compactions even if no duplicate keys are found
  Now, if no duplicate keys are identified after all messages are collected from all involved RO Logs, no compaction takes place.

It's probably a good idea to proceed with compaction if it's going to reduce the number of tracked RO Logs. Currently, after we have parsed the messages from all RO Logs, we generate a new ROLog for each input ROLog, as opposed to building bundles from potentially multiple previous segments.

• Spill buffers for (content, keys) to disk when parsing messages in order to be able to process more messages than be held in memory
  We 'd need to mmap() those on-disk files later and dereference them directly in the VMA