fix: remove ix_moe_bridge — nm -D confirms libixformer.so has NO MoE symbols

真机探测确认:
  nm -D libixformer.so | grep topk_softmax → 空
  ixf_F dir() → 无 vllm_moe_topk_softmax
  ixf_F dir() → 无 vllm_invoke_fused_moe_kernel
  ixf_F dir() → 无 vllm_moe_align_block_size
  _ixformer_torch.so symbols → 仅 cuinfer_gemm 系列, 无 MoE

结论: base 镜像的 MoE 路径:
  fused_moe.py → _custom_ops.topk_softmax → ixf_F.vllm_moe_topk_softmax → AttributeError
  → qwen3_5.py 捕获 → fallback to Python expert loop (这是唯一能工作的路径)

修改:
1. _custom_ops.py topk_softmax: 直接 PyTorch softmax+topk, 不尝试 ixf_F (消除 ERROR 日志)
2. 移除 ix_moe_bridge 加载逻辑 (libixformer.so 没有 MoE 符号, 链接会失败)
3. 移除 patch_ops.sh ix_moe_bridge JIT 编译步骤

comp 168 的 0 分根因不是 MoE fallback (所有参赛者都 fallback),
而是我们的自定义 qwen3_5.py 导致 GDN NaN 99.98% + OOM.
上一个 commit 已修复: 条件部署 qwen3_5.py + max_model_len=80000.

qwen3_6_scripts/patch_ops.sh

@@ -260,37 +260,7 @@ if [ -d "$EX_ENGINE_SRC/python" ]; then
     echo "[patch_ops] EX Engine Python package deployed to $EX_PY_DST"
 fi
 
-# 7a. JIT compile ix_moe_bridge.cpp → .so (bridge to ixformer::infer C++ API)
-# This is CRITICAL: topk_softmax, group_gemm, etc. are ONLY accessible via C++
-IX_MOE_BRIDGE_CPP="/workspace/ex_engine/csrc/ix_moe_bridge.cpp"
-if [ -f "$IX_MOE_BRIDGE_CPP" ]; then
-    echo "[patch_ops] Pre-compiling ix_moe_bridge.cpp (ixformer C++ bridge)..."
-    python3 -c "
-import torch
-from torch.utils.cpp_extension import load
-try:
-    bridge = load(
-        name='ix_moe_bridge',
-        sources=['$IX_MOE_BRIDGE_CPP'],
-        extra_include_paths=['/usr/local/corex/include'],
-        extra_ldflags=[
-            '-L/usr/local/corex/lib64',
-            '-L/usr/local/corex/lib64/python3/dist-packages/ixformer',
-            '-lixformer',
-            '-Wl,-rpath,/usr/local/corex/lib64/python3/dist-packages/ixformer',
-        ],
-        verbose=True,
-    )
-    print('[patch_ops] ix_moe_bridge compiled successfully')
-    # Test basic function availability
-    print(f'[patch_ops] Bridge functions: {[x for x in dir(bridge) if not x.startswith(\"_\")]}')
-except Exception as e:
-    print(f'[patch_ops] WARNING: ix_moe_bridge compile failed: {e}')
-    print('[patch_ops] topk_softmax will use PyTorch fallback')
-" 2>&1 || echo "[patch_ops] WARNING: ix_moe_bridge pre-compile step failed"
-fi
-
-# 7b. Precompile MoE topk_softmax CUDA kernel (.cu → .so)
+# 7. Precompile MoE topk_softmax CUDA kernel (.cu → .so)
 # This replaces the missing ixf_F.vllm_moe_topk_softmax with our own CUDA kernel
 MOE_TOPK_CU="/workspace/ex_engine/csrc/moe_topk_softmax_v3.cu"
 if [ -f "$MOE_TOPK_CU" ]; then