[Spec] Refactor NaN/OOB checks to async maybe_detect_* with env-var control (#19899)

Co-authored-by: hnyls2002 <lsyincs@gmail.com>

test/registered/spec/eagle/test_deepseek_v3_fp4_mtp_small.py

@@ -50,7 +50,13 @@ class TestDeepseekV3FP4MTP(CustomTestCase):
             "--model-loader-extra-config",
             '{"enable_multithread_load": true,"num_threads": 64}',
         ]
-        with envs.SGLANG_ENABLE_SPEC_V2.override(True):
+        with envs.SGLANG_ENABLE_SPEC_V2.override(
+            True
+        ), envs.SGLANG_SPEC_NAN_DETECTION.override(
+            True
+        ), envs.SGLANG_SPEC_OOB_DETECTION.override(
+            True
+        ):
             cls.process = popen_launch_server(
                 cls.model,
                 cls.base_url,

test/registered/spec/eagle/test_eagle_constrained_decoding.py

@@ -59,7 +59,13 @@ class TestEagleConstrainedDecoding(
             cls.grammar_backend,
         ]
         launch_args.extend(cls.other_launch_args)
-        with envs.SGLANG_ENABLE_SPEC_V2.override(cls.spec_v2):
+        with envs.SGLANG_ENABLE_SPEC_V2.override(
+            cls.spec_v2
+        ), envs.SGLANG_SPEC_NAN_DETECTION.override(
+            True
+        ), envs.SGLANG_SPEC_OOB_DETECTION.override(
+            True
+        ):
             cls.process = popen_launch_server(
                 cls.model,
                 cls.base_url,

test/registered/spec/eagle/test_eagle_dp_attention.py

@@ -3,6 +3,7 @@ from types import SimpleNamespace
 
 import requests
 
+from sglang.srt.environ import envs
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.few_shot_gsm8k import run_eval as run_eval_few_shot_gsm8k
 from sglang.test.send_one import BenchArgs, send_one_prompt
@@ -55,12 +56,15 @@ class TestEAGLE3EngineDPAttention(CustomTestCase):
             "--cuda-graph-max-bs",
             "64",
         ]
-        cls.process = popen_launch_server(
-            cls.model,
-            cls.base_url,
-            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
-            other_args=other_args,
-        )
+        with envs.SGLANG_SPEC_NAN_DETECTION.override(
+            True
+        ), envs.SGLANG_SPEC_OOB_DETECTION.override(True):
+            cls.process = popen_launch_server(
+                cls.model,
+                cls.base_url,
+                timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+                other_args=other_args,
+            )
 
     @classmethod
     def tearDownClass(cls):

test/registered/spec/eagle/test_eagle_infer_beta.py

@@ -59,7 +59,13 @@ class TestEagleServerBase(CustomTestCase, MatchedStopMixin):
         launch_args.extend(cls.other_launch_args)
         with envs.SGLANG_ENABLE_SPEC_V2.override(
             True
-        ), envs.SGLANG_ENABLE_STRICT_MEM_CHECK_DURING_BUSY.override(1):
+        ), envs.SGLANG_ENABLE_STRICT_MEM_CHECK_DURING_BUSY.override(
+            1
+        ), envs.SGLANG_SPEC_NAN_DETECTION.override(
+            True
+        ), envs.SGLANG_SPEC_OOB_DETECTION.override(
+            True
+        ):
             cls.process = popen_launch_server(
                 cls.model,
                 cls.base_url,

test/registered/spec/eagle/test_eagle_infer_beta_dp_attention.py

@@ -65,7 +65,13 @@ class TestEagleDPAttnServerSmall(CustomTestCase):
             "--speculative-num-draft-tokens",
             "4",
         ]
-        with envs.SGLANG_ENABLE_SPEC_V2.override(True):
+        with envs.SGLANG_ENABLE_SPEC_V2.override(
+            True
+        ), envs.SGLANG_SPEC_NAN_DETECTION.override(
+            True
+        ), envs.SGLANG_SPEC_OOB_DETECTION.override(
+            True
+        ):
             cls.process = popen_launch_server(
                 cls.model,
                 cls.base_url,

test/registered/spec/eagle/test_eagle_infer_beta_dp_attention_large.py

@@ -73,7 +73,13 @@ class TestEagleDPAttnServerLarge(CustomTestCase):
             "--model-loader-extra-config",
             '{"enable_multithread_load": true,"num_threads": 64}',
         ]
-        with envs.SGLANG_ENABLE_SPEC_V2.override(True):
+        with envs.SGLANG_ENABLE_SPEC_V2.override(
+            True
+        ), envs.SGLANG_SPEC_NAN_DETECTION.override(
+            True
+        ), envs.SGLANG_SPEC_OOB_DETECTION.override(
+            True
+        ):
             cls.process = popen_launch_server(
                 cls.model,
                 cls.base_url,

test/registered/spec/test_constrained_decoding_spec_reasoning.py

@@ -3,6 +3,7 @@ import unittest
 
 import openai
 
+from sglang.srt.environ import envs
 from sglang.srt.utils import kill_process_tree
 from sglang.test.ci.ci_register import register_cuda_ci
 from sglang.test.test_utils import (
@@ -50,12 +51,15 @@ class ServerWithGrammar(CustomTestCase):
             "--speculative-num-draft-tokens=8",
         ]
 
-        cls.process = popen_launch_server(
-            cls.model,
-            cls.base_url,
-            timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
-            other_args=launch_args,
-        )
+        with envs.SGLANG_SPEC_NAN_DETECTION.override(
+            True
+        ), envs.SGLANG_SPEC_OOB_DETECTION.override(True):
+            cls.process = popen_launch_server(
+                cls.model,
+                cls.base_url,
+                timeout=DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
+                other_args=launch_args,
+            )
 
     @classmethod
     def tearDownClass(cls):