...because the latter lacks the abs(long) overload in some popular environments,
cf. <http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60401> "stdlib.h does not
provide abs(long) overload."

Similarly, stdlib.h lacks the abs(float), abs(double), abs(long double)
overloads compared to cmath there, whose use was apparently intended in
sw/qa/extras/ooxmlexport/ooxmlsdrexport.cxx.  Rewrote that to use
CPPUNIT_ASSERT_DOUBLES_EQUAL instead, which revealed that the comparisons need
rather large deltas of .1 resp. .2 (which the original code hid with an
implicit conversion to integral type, thus using an effective delta of 1).

Discovered with -Wabsolute-value ("absolute value function 'abs' given an
argument of type 'long' but has parameter of type 'int' which may cause
truncation of value") recently introduced on Clang trunk towards 3.5.

Change-Id: I4c41575ffdccb2944498b662bd3a53fd510cb0c0