Fast and drity (regex-based) XCES parser based on the one in Pantera, noticeably faster but prone to breakage on even slightly freeform xml files

initial switch to use shared_ptrs for sentence and chunk objects, ale get rid of the pointless sentence template

In case the xces document contains sentences with no chunk containing them, a fake chunk will be created to hold the sentences.
Previously this situation threw an exception.

Libcorpus2 now contains the base strructures (tokens etc), their I/O and related classes.
Libpwrutils (name TBD) contains basic building blocks common to all tagger-related libraries
Tagset-tool only needs libcorpus, so was brough here as well.