Performance: While iterating buckets, use batch reads

[lvca]

Today's session with the profiler showed an interesting bottleneck with scans/iteration of types/buckets: the cost of acquiring the database read lock is higher than expected for such a simple operation (read a record).

In some cases, like the scan of a bucket/type, it would be nice to have a batch of reads all executed in the same lock. This would keep the read operation safe (against a backup, for example) and it would speed up the queries.