removed ddepth from matrix (after Vadim's commit, it generates problem with core and ml tests)

returned the new floating-point number formatting

modules/core/src/matrix.cpp

@@ -1873,7 +1873,7 @@ void cv::reduce(const InputArray& _src, OutputArray _dst, int dim, int op, int d
         op = CV_REDUCE_SUM;
         if( sdepth < CV_32S && ddepth < CV_32S )
             temp.create(dst.rows, dst.cols, CV_32SC(cn));
-        ddepth = CV_32S;
+        //ddepth = CV_32S;
     }
 
     ReduceFunc func = 0;