* xmlreader.c: seriously changed the way data are pushed to
  the underlying parser, go by block of 512 bytes instead of
  tryng to detect tag boundaries at that level. Changed the
  way empty element are detected and tagged.
* python/tests/reader.py python/tests/reader2.py
  python/tests/reader3.py: small changes mostly due to context
  reporting being different and DTD node being reported. Some
  errors previously undetected are now caught and fixed.
* doc/xmlreader.html: flagged last section as TODO
Daniel