Remove Fsyncs from File Info Query

When requesting large numbers of files at once, we used to perform a blocking writeaheadlog update for each file that we scanned.

This MR reorganizes the code a bit, so that when requesting large numbers of files in the renter it is only doing reads, not writes.

This results in a ~20x speedup for requesting a large number of files. There are still performance issues for requesting large numbers of files at once, but I think that will be mostly mitigated after doing per-directory lookups.