[Fix] Fix gpt oss triton kernels and upgrade flashinfer back to 0.6.11.post1 (#25335)

Co-authored-by: sglang-bot <sglang-bot@users.noreply.github.com>
Co-authored-by: b8zhong <b8zhong@users.noreply.github.com>
Co-authored-by: mmangkad <mmangkad@users.noreply.github.com>

test/registered/unit/layers/quantization/test_mxfp4_sm90_cutlass.py

@@ -20,12 +20,7 @@ import torch
 
 from sglang.test.ci.ci_register import register_cuda_ci
 
-register_cuda_ci(
-    est_time=120,
-    stage="stage-b",
-    runner_config="1-gpu-large",
-    disabled="broken on main, see #24816",
-)
+register_cuda_ci(est_time=120, stage="stage-b", runner_config="1-gpu-large")
 
 flashinfer_fused_moe = pytest.importorskip("flashinfer.fused_moe")