support cuda_graph on iluvatar (#232)

### PR Category
Others

### PR Type
Bug Fixes 

### Description
support cuda_graph on iluvatar，Has been verified Qwen3.6-27B and
Qwen3.6-35B-A3B models
test command:
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
export VLLM_ENGINE_ITERATION_TIMEOUT_S=72000
export VLLM_RPC_TIMEOUT=72000000
export VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS=7200
vllm serve /models/Qwen3.6-35B-A3B/ -tp 8 --served-model-name qwen
--port 8070 --max-model-len 131072 --compilation-config
'{"cudagraph_mode": "FULL_DECODE_ONLY"}'

Note: If use the default `compilation-config` parameter, maybe OOM.

### Related Issues
<!-- Link any related issues: Fixes #issue, Closes #issue, or Related to
#issue -->

### Changes
<!-- List the key changes made in this PR. -->
-

### Testing
<!-- How has this change been tested? Include test commands, hardware
used, etc. -->
-

### Checklist
- [ ] I have run the existing tests and they pass
- [ ] I have added tests for my changes (if applicable)
- [ ] I have updated the documentation (if applicable)

Author: luoyc123

vllm_fl/platform.py

@@ -333,7 +333,7 @@ def get_static_graph_wrapper_cls(cls) -> str:
 
     @classmethod
     def support_static_graph_mode(cls) -> bool:
-        if cls.vendor_name in ["nvidia", "ascend", "metax", "hygon", "mthreads"]:
+        if cls.vendor_name in ["nvidia", "ascend", "metax", "hygon", "mthreads", "iluvatar"]:
             return True
         return False
 
@@ -414,7 +414,7 @@ def get_device_total_memory(cls, device_id: int = 0) -> int:
 
     @classmethod
     def use_custom_op_collectives(cls) -> bool:
-        return cls.vendor_name in ("nvidia", "thead")
+        return cls.vendor_name in ("nvidia", "thead", "iluvatar")
 
     @classmethod
     def num_compute_units(cls, device_id: int = 0) -> int: