Fix MODEL_CONVERTER_LIB_DIR path resolution for VGF tests (pytorch#20907)

Summary:

After bumping model-converter to v0.9.0 (D109721412), ~160 ARM VGF tests
(mxfp_linear, mxfp_conv2d) started SKIPPING because model_converter_installed()
returns False on sandcastle.

Root cause: MODEL_CONVERTER_LIB_DIR was set to
  $(location fbsource//third-party/nvidia-nsight-systems:linux-x86_64)/host-linux-x64
which relies on string concatenation after $(location) expansion. In Buck2 test
sandboxes, this pattern can produce an invalid path because the export_file
artifact reference may not expose subdirectories properly.

Fix: Create a dedicated export_file target for the host-linux-x64 subdirectory
and reference it directly with $(location), eliminating the string concatenation.

Impact: −160 SKIPPING tests

Reviewed By: rascani

Differential Revision: D111819484

Author: psiddh

backends/arm/test/targets.bzl

@@ -102,7 +102,7 @@ def define_arm_tests():
         if not runtime.is_oss and _ENABLE_VGF:
             test_env.update({
                 "MODEL_CONVERTER_PATH": "$(location fbsource//third-party/pypi/ai-ml-sdk-model-converter/0.9.0:model-converter-bin)",
-                "MODEL_CONVERTER_LIB_DIR": "$(location fbsource//third-party/nvidia-nsight-systems:linux-x86_64)/host-linux-x64",
+                "MODEL_CONVERTER_LIB_DIR": "$(location fbsource//third-party/nvidia-nsight-systems:host-linux-x64-libs)",
                 "LAVAPIPE_LIB_PATH": "$(location fbsource//third-party/mesa:vulkan_lvp)",
                 "EMULATION_LAYER_TENSOR_SO": "$(location fbsource//third-party/arm-ml-emulation-layer/v0.9.0/src:libVkLayer_Tensor)",
                 "EMULATION_LAYER_GRAPH_SO": "$(location fbsource//third-party/arm-ml-emulation-layer/v0.9.0/src:libVkLayer_Graph)",