Big character set cleanup

I just tried compiling Conglomerate under GCC4, which is a lot more
picky about the signedness of character types (and pointers to these
types) than GCC3.

There were _lots_ of warnings due to conversions between xmlChar* and
gchar* types.  

Both xmlChar* and gchar* are pointers to zero-terminated, UTF-8 encoded
strings, so they're readily interchangable when passed as const - though
they have different allocation/deallocation functions, so you have to be
careful if passing them in a non-const function (i.e. to a function that
might free or assume ownership of the memory).

Previously we were using a CongXMLChar type, which was never very
clearly designed.   Since we're a GTK app I think it's fair enough to
use gchar throughout for representing UTF-8 encoded strings, apart from
the final interface with libxml2 and libxslt.

So I've gone through and purged the CongXMLChar type, and added the
necessary casts and conversion to/from xmlChar where we interface with
the libxml2 types  (this was quite a bit of work; I fixed some uses of
the wrong deallocation function as I went along).

CVS conglomerate now compiles cleanly with GCC4 with -Wall -Werror

I also got renamed the function xml_frag_data_nice to
cong_node_safe_get_content (this was a long-standing survivor from the
original conglomerate prototype from 1999!)

It'd be great if people using different compilers (e.g. GCC 3) could
check out the latest code and confirm that it compiles cleanly with the
highest warning coverage.  It'd also be good if you could grab the code
anyway, and test it; I hope I haven't introduced any new bugs, but it
was a lot of typing.... 

Thanks!

Dave