HvD: Updated the test case a bit. The convergence on the geometry optimization

is normally detected on one of two things. Either the gradient and the step
size are both smaller than a threshold, OR the gradient is much smaller than
the threshold (in which case the step size is ignored). This causes variations
of one step in the geometry optimization depending on whether the gradient
accidentily happens to be small or really small. These variations cause 
spurious test case failures. Therefore I have now selected lacks step size
settings so that calculation converges on the gradient alone in all cases.

QA/tests/cosmo_h3co/cosmo_h3co.nw

@@ -22,6 +22,8 @@ end
 
 driver
   tight
+  xmax 0.00180
+  xrms 0.00120
 end
 
 task scf optimize