WiktionarySplitter: Support compressed inputs.

WiktionarySplitter: Support compressed inputs.

Unfortunately bzip2 decompression is very slow (slower
than the XML parsing in fact), so it might make sense to
re-compress the downloaded files from bzip2 to xz.
If the decompression could be done in a separate thread,
xz compression would even provide a speedup if the files
are on a slower (non-SSD) disk.