More efficient node traversal during garbage collection

During the 'marking' stage the data store janitor runs over all nodes in the repository and updates the last modified date of any binary that is stored in the data store table that it finds. However, this process loads all node states in the repository individually which means that a separate database select is run for every node in the repository. This is very inefficient. We can improve on this manyfold by using the same strategy as is used by the consistency checker: we load batches of node info objects at once that already contain enough information to do our processing.