[MERGE] res.partner: current name_search() implementation tends to skip a few valid...

The main reason for the semi-random behavior
observed during auto-completion is the
missing ORDER BY clause in the pre-filtering
SQL query.

The ORDER BY clause is expensive but inevitable
if we want to apply a correct LIMIT, otherwise
we would return random `limit` results among
all the possible matches.

The current SQL query seems convoluted due
to the duplicated CASE clause but it
performs slightly better than the equivalent
CTE-based (WITH...) query, so it was preferred.

There is still a chance of returning too
few results due to double limit application,
as further discussed in bug 1203727

lp bug: https://launchpad.net/bugs/1203727 fixed

bzr revid: odo@openerp.com-20130906134950-gi0szic3uw3onyuv