[https://nvbugs/5604136][fix] AutoDeploy: correct import for mxfp4_moe unit test (#8593)

Signed-off-by: Lucas Liebenwein <[email protected]>

Author: lucaslie

tests/unittest/_torch/auto_deploy/unit/multigpu/custom_ops/test_mxfp4_moe_ep.py

@@ -5,12 +5,11 @@
 import torch.distributed as dist
 from _dist_test_utils import get_device_counts
 
+from tensorrt_llm._torch.auto_deploy.custom_ops.fused_moe.mxfp4_moe import (
+    IS_TRITON_KERNELS_AVAILABLE,
+)
 from tensorrt_llm._torch.auto_deploy.distributed.common import spawn_multiprocess_job
 
-# FIXME: https://nvbugspro.nvidia.com/bug/5604136.
-# from tensorrt_llm._torch.auto_deploy.custom_ops.mxfp4_moe import IS_TRITON_KERNELS_AVAILABLE
-IS_TRITON_KERNELS_AVAILABLE = False
-
 
 def _split_range_last_remainder(n: int, world_size: int, rank: int):
     """[lo, hi) split along dim0; last rank gets remainder."""