All tags are inserted as Tag::Other, and no matchup to UMTHES is done.

A single record can have multiple languages attached (we now just use the first
one) and it might not use a language code at all.

We currently have 22k directories containing 1.2 files in the mean.

Since this is inefficient, this commit reduces the part of the hash
used to determine the directory, i.e. there will be fewer directories
with more files in them.

Parse information in CSW harvester.

We only use response replay during development now where the additional
compression is a waste of CPU time as we have ample disk space and usually work
with a limited set of harvesters in any case.