[CI] Relax gpt-oss 4GPU accuracy threshold from 0.60 to 0.58 (#22237)

Co-authored-by: Alison Shao <alison.shao@Mac.attlocal.net>

test/registered/4-gpu-models/test_gpt_oss_4gpu.py

@@ -13,7 +13,7 @@ class TestGptOss4Gpu(BaseTestGptOss):
             model_variant="120b",
             quantization="bf16",
             expected_score_of_reasoning_effort={
-                "low": 0.60,
+                "low": 0.58,
             },
             other_args=["--tp", "4", "--cuda-graph-max-bs", "200"],
         )
@@ -23,7 +23,7 @@ class TestGptOss4Gpu(BaseTestGptOss):
             model_variant="120b",
             quantization="mxfp4",
             expected_score_of_reasoning_effort={
-                "low": 0.60,
+                "low": 0.58,
             },
             other_args=[
                 "--tp",