Logs deletion support for boltdb-shipper store

[sandeepsukhani]

What this PR does / why we need it:
This PR leverages existing code for retention to support deletion for boltdb-shipper.
It relies on the Mark phase of retention which now also looks at eligible delete requests by talking to DeleteRequestsManager to decide whether the chunk is supposed to be deleted or not. For partially deleted chunks I have added a chunk rewriter which would build chunks for the interval for which they are not deleted and writes them to the store. A successful Mark process over all the tables considers delete requests as processed and the relevant chunk should be deleted later by the Sweep process.

The delete requests are persisted in a table called delete_requests which is a boltdb file. The file is uploaded to the object store every 5 mins or during shutdown and is downloaded locally if it does not exist during startup.

Here is how deletion would roughly work:

1. Call MarkPhaseStarted of DeleteRequestsManager before starting the compaction which would load eligible delete requests in memory.
2. Each Mark phase of the table would check if one or more delete requests deletes the chunks in it and take care of deleting the chunk index and rewriting it if it was partially deleted.
3. Call MarkPhaseFinished after successful compaction of all the tables which would mark the in-memory requests as processed.
4. Call MarkPhaseFailed if any of the tables fails to compact which would just drop requests from memory.

Checklist

• Documentation added
• Tests updated

[cyriltovena]

LGTM.

I'd appreciate if you could grab the userID without allocation, when deleting millions of chunks it might slow everything down.