Arm backend: Remove _vgf_ op pytest xfails (pytorch#14116)

Use VK_LAYER_PATH instead of VK_ADD_LAYER_PATH when setting up mlsdk
deps since looks like some loaders ignore VK_ADD_LAYER_PATH.

Author: mansnils

backends/arm/scripts/mlsdk_utils.sh

@@ -145,7 +145,7 @@ function setup_path_emulation_layer() {
     prepend_env_in_setup_path DYLD_LIBRARY_PATH "${model_emulation_layer_path}/deploy/lib"
     prepend_env_in_setup_path VK_INSTANCE_LAYERS VK_LAYER_ML_Tensor_Emulation
     prepend_env_in_setup_path VK_INSTANCE_LAYERS VK_LAYER_ML_Graph_Emulation
-    prepend_env_in_setup_path VK_ADD_LAYER_PATH "${model_emulation_layer_path}/deploy/share/vulkan/explicit_layer.d"
+    prepend_env_in_setup_path VK_LAYER_PATH "${model_emulation_layer_path}/deploy/share/vulkan/explicit_layer.d"
 }
 
 #setup_model_converter() $1

backends/arm/test/ops/test_add.py

@@ -207,7 +207,6 @@ def test_add_tensor_u85_INT_2(test_data: input_t2):
 
 @common.parametrize("test_data", filtered_test_data)
 @common.SkipIfNoModelConverter
-@pytest.mark.xfail(reason="'Failed to load VKML extensions' error in ci.")
 def test_add_tensor_vgf_FP(test_data: input_t1):
     pipeline = VgfPipeline[input_t1](
         Add(),
@@ -225,7 +224,6 @@ def test_add_tensor_vgf_FP(test_data: input_t1):
 
 @common.parametrize("test_data", filtered_test_data)
 @common.SkipIfNoModelConverter
-@pytest.mark.xfail(reason="'Failed to load VKML extensions' error in ci.")
 def test_add_tensor_vgf_INT(test_data: input_t1):
     pipeline = VgfPipeline[input_t1](
         Add(),