Clean log lines immediately when they are read and also make use of sanitized
log's high redundancy immediately, i.e., continue with maps of
<LocalDate, <Map<String, Long>>.

Rename method(s) to reflect what they do.

Adapt to latest changes of metrics-lib (task-25329) and make use of the high
redundancy of logs (e.g. a 3G file might only contain 350 different lines).
This avoids OOM and array out of bounds exceptions for large files (>2G) and
gives a speed-up of roughly 50%. (The earlier 66min are down to 34min for
meronense&weschniakowii files plus two larger files.)

There is a BATCH constant, which could be tuned for processing speed. It is
logged for each webstats module run.  Currently, it is set to 100k.  This
was more or less arbitrarily chosen and used for all the tests.  A test run
using 500k didn't show significant differences.

Also edit here and there for more readability and less lines.

Add convenience constructor accepting the first string as argument.

To avoid possible inconsistencies DescriptorBuilder is finalized after the first
call to 'toString' and cannot be altered anymore.  Any attempt to add more leads
to an IllegalStateException.

This might facilitate easier processing of descriptors.