Measure db bucket length can take a long time / stall

On occasion the "Visor init" section will stall or take a long time; the only apparent cause of this is the bucket `Len` measurement. It is likely stalling on the `bkt.Stats` call. Review boltdb docs, consider coreos/bbolt and if necessary we'll need to count the bucket size manually by maintaining a counter.

This stall has appeared on daemons on remote servers and is interfering with deployments and upgrades.

Start at the Visor init section that measures a bucket with bkt.Len and investigate the bkt.Stats call. Review the boltdb documentation and coreos/bbolt behavior, then verify that bucket-length measurement no longer stalls remote daemons while still reporting the correct size.

Written by the indexing model from the issue text.