Compact db

[vincent-petithory]

This adds a compact to the bolt tool. It recursively inspects a bolt database and copies its keys to a new database.
This usually results in a compacted database, especially if there was a lot of deleted keys with large values. See #308

Thanks for taking a look at this!

[jacek99]

This is moderately useful, but forces downtime as the app has to go down and be restarted pointing to a new DB.

It would be great if this could be done on the current DB file while it is being used, to enable high availability (similar to the way Cassandra does compaction while staying online all the time).

[chrsm]

See also #423

[mdlayher]

Cool stuff, would love to see this integrated into the tool, at least.

Online compaction would be great, but I certainly think this is a nice feature to have regardless.

[tg]

Just run this on my 7GB db file and it was successfully truncated to 2GB. Happy days! A few comments:

1. I think source DB should be open in read only mode, otherwise source DB is being modified (when testing on some 167MB sample DB I actually seen the source DB being inflated to 256MB). Although this is an issue with all bolt CLI commands like check or stats, so maybe should be a separate issue.
2. I would suggest output DB should share the same file permissions as source DB rather then 0666. We already call os.Stat(path) so this as simple as taking Mode().Perm() from the result and passing to bolt.Open.
3. Would it make sense to have a sensible non-zero default for -tx-max-size? I bet this would save a few souls from killing their machines while running compaction on very large DBs without reading help carefully. If someone would like to optimise the process using a single transaction, they would need to do it explicitly.
4. As I mentioned in Question: Shrink BoltDB File? #423 (comment), this compacting approach doesn't copy bucket sequence number, but as far as I can see there is no solution to this using the public interface. I guess most people don't use this functionality anyway, so we should at least add some warning in the help message or maybe somehow be able to detect buckets with non-zero sequence numbers and fail the compaction.

[benbjohnson]

Sorry for the ridiculously long delay. I posted a pull request which includes the commits from this PR as well as some additional features for copying buckets sequence. The new pull request is #590.