Allow module-level skip in comm GEMM overlap tests

Author: almogsegal

tests/pytorch/distributed/test_comm_gemm_overlap.py

@@ -11,7 +11,7 @@
 import transformer_engine.pytorch.cpp_extensions as tex
 
 if torch.cuda.device_count() < 2:
-    pytest.skip("Comm+GEMM overlap requires at least 2 GPUs.")
+    pytest.skip("Comm+GEMM overlap requires at least 2 GPUs.", allow_module_level=True)
 
 fp8_available, reason_for_no_fp8 = te.is_fp8_available(return_reason=True)
 mxfp8_available, reason_for_no_mxfp8 = te.is_mxfp8_available(return_reason=True)