Behavior of apr_atomic_add32

[Jim Jagielski]

I've added a test to APR-1.5 and trunk so see if apr_atomic_add32
acts "as expected" when adding a negative number...