Made loss layers output the gradients by assigning them to the output rather

than adding them.  This way, the gradient buffer can be used as scratch space
during the loss computation.

dlib/dnn/loss.h

@@ -77,7 +77,7 @@ namespace dlib
                 if (temp > 0)
                 {
                     loss += scale*temp;
-                    g[i] += -scale*y;
+                    g[i] = -scale*y;
                 }
             }
             return loss;

dlib/dnn/loss_abstract.h

@@ -110,9 +110,9 @@ namespace dlib
                   of sub matches the expected labels given by truth.  Let's write the loss
                   function as L(input_tensor, truth, sub).  
                 - Then compute_loss() computes the gradient of L() with respect to the
-                  outputs in sub.  Specifically, compute_loss() adds the gradients into sub
+                  outputs in sub.  Specifically, compute_loss() assigns the gradients into
-                  by performing the following tensor additions, for all valid i: 
+                  sub by performing the following tensor assignments, for all valid i: 
-                    - layer<i>(sub).get_gradient_input() += the gradient of
+                    - layer<i>(sub).get_gradient_input() = the gradient of
                       L(input_tensor,truth,sub) with respect to layer<i>(sub).get_output().
                 - returns L(input_tensor,truth,sub)
         !*/