[SYCL] Avoid re-computing group_range in nd_item class

This fix is mostly NFC.
Instead of doing costly division operations and re-computing the group_range
inside nd_item class, it is better to call group::get_group_range() that
does not do divisions as the group class keeps the group_range
as a pre-computed field.

Signed-off-by: Vyacheslav N Klochkov <vyacheslav.n.klochkov@intel.com>

Author: v-klochkov

sycl/include/CL/sycl/nd_item.hpp

@@ -82,12 +82,11 @@ template <int dimensions = 1> class nd_item {
   }
 
   range<dimensions> get_group_range() const {
-    return Group.get_global_range() / Group.get_local_range();
+    return Group.get_group_range();
   }
 
   size_t __SYCL_ALWAYS_INLINE get_group_range(int dimension) const {
-    size_t Range =
-        Group.get_global_range(dimension) / Group.get_local_range(dimension);
+    size_t Range = Group.get_group_range(dimension);
     __SYCL_ASSUME_INT(Range);
     return Range;
   }