[ET-VK] Replacing use of adaptive_work_group_size function by create_local_wg_size function for slice and copy ops.

Pull Request resolved: #7016

This diff replaces the use of the adaptive_work_group_size function with create_local_wg_size function for slice and copy ops, which is better tuned for improving shader performance.
ghstack-source-id: 256094684
@exported-using-ghexport

Differential Revision: [D66308779](https://our.internmc.facebook.com/intern/diff/D66308779/)

Author: trivedivivek

backends/vulkan/runtime/graph/ops/impl/Copy.cpp

@@ -135,7 +135,7 @@ void add_copy_channel_offset_node(
         utils::safe_downcast<uint32_t>(dim_at<kWidth4D>(in_sizes)),
         utils::safe_downcast<uint32_t>(dim_at<kHeight4D>(in_sizes)),
         utils::safe_downcast<uint32_t>(dst_last_z - dst_first_z + 1)};
-    uvec3 local_size = adaptive_work_group_size(global_size);
+    uvec3 local_size = graph.create_local_wg_size(global_size);
 
     const struct Block final {
       ivec3 range;

backends/vulkan/runtime/graph/ops/impl/Slice.cpp

@@ -126,7 +126,7 @@ void add_slice_tensor_copy_node(
     add_dtype_suffix(kernel_name, *t_out);
 
     utils::uvec3 global_size = t_out->logical_limits();
-    utils::uvec3 local_size = adaptive_work_group_size(global_size);
+    utils::uvec3 local_size = graph.create_local_wg_size(global_size);
 
     const struct Block final {
       int dim;