fix(build): 回退qwen3_6_scripts+ex_engine到26e6cb40(能得分版本)

唯一改动: computility-run.yaml max_model_len 80000→100000

26e6cb40是Sub520能在竞赛平台docker build成功并得分的版本
之后所有commit都导致docker build失败
根因: 新增的65个文件(vendor_overrides/prebuilt/*.so/wheels等)
可能触发了竞赛平台docker build的某个限制

本次回退:
- qwen3_6_scripts/: 110→45文件(删掉65个新增文件)
- ex_engine/: 恢复到26e6cb40完全一致
- Dockerfile: 恢复5个RUN步骤结构(已验证能build)
- computility-run.yaml: max_model_len=100000(避免replay 400拒绝)

qwen3_6_scripts/xformers.py

@@ -942,10 +942,18 @@ class XFormersImpl(AttentionImpl[XFormersMetadata]):
             query = query.unsqueeze(0)
             key = key.unsqueeze(0)
             value = value.unsqueeze(0)
-            # BI-V100: ixformer varlen_fwd has incompatible signature with
-            # xops.fmha.flash.FwOp() (20-arg mismatch). Use pure-math fallback
-            # for ALL head sizes during prefill, not just head_size > 128.
-            out = self._run_sdpa_fallback(query, key, value, attn_metadata)
+            if self.head_size > 128:
+                out = self._run_sdpa_fallback(query, key, value, attn_metadata)
+            else:
+                out = xops.memory_efficient_attention_forward(
+                    query,
+                    key,
+                    value,
+                    attn_bias=attn_bias[0],
+                    p=0.0,
+                    scale=self.scale,
+                    op=self.attn_op,
+                )
             return out.view_as(original_query)
 
         # Attention with alibi slopes.