xpath improvement V2

Hi all,

I've combined the review suggestions of Jan Urbański, Scott Bailey, and
others.
This was a lot harder, then I had foreseen; and I took my time to do it
the right way (hope you agree!).

BTW. really appreciate your efforts, so far, to enlighten me on nuub
errors/mistakes in the previous version.

Additional improvement (hence the V2):
two extra functions: xpath_value_text and xpath_value_strict
Both are quite general: xpath_value_text maps everything to text, except
nodeset. xpath_value_strict has to be told exactly what to expect.
xpath_value_text(text xpath, xml doc [, namespace]) returns text
xpath_value_strict(anyelement typexample, text xpath, xml doc [,
namespace]) returns anyelement
See the doc for further explanation.
I chose this approach, as opposed to xpath_value_string/number/boolean
for a couple of reasons:
- We want postgresql functions, with postgresql types. The functions
should fit postgresql usage and hide libxml parlance.
- Functions in pg_catalog should be destined for broad use, not just
satisfy the adhoc desire of the implementer.
- Function synopsis should be adequate to withstand libxml extension or
improvements (libxml3?)
- Construction of XPath expressions for value retrieval is not trivial.
A precise calling syntax, hopefully, focuses the user to select the
right expressions.
- Loose implementation with "autocasting" opens the door for unwanted
injection possibilities.

Lastly, when in need of a xpath_value_string (et al), it can easily be
added by the user through:
CREATE FUNCTION xpath_value_string(text, xml) RETURNS text AS $$
SELECT xpath_value_strict('a'::text,$1,$2);
$$ LANGUAGE SQL;

Points fixed:
- source code indentation. (manually though, can't get pgindent to work
properly)
- Doc entries with some examples
- test/regress entries
- Detailed directions from Jan Urbanski on ereport and PG_TRY/CATCH.

Here's a shortlist of subjects you should definetely review (aka I'me
not certain these are up to standards)
- calling style in xpath_value_strict with typexample. I really like it,
but I haven't seen such an approach elsewhere in the API.
- tone and style in doc. I'me not native English speaker and had to
learn DocBook along the way.
- insertion in catalog/pg_proc.h. Just took some free oid numbers, and
fiddled the other parameters.
- Is ERRCODE_DATA_EXCEPTION (in xml.c) acceptable as error code for a
type mismatch? Could not find anything better.

kind regards, Arie Bikker