Enable async-assert invariant probes by default in CI (#27461)

.github/workflows/_pr-test-stage.yml

@@ -60,6 +60,7 @@ on:
 # defines must be redeclared here for the called job to see the same context.
 env:
   SGLANG_IS_IN_CI: true
+  SGLANG_ENABLE_ASYNC_ASSERT: true
   SGLANG_CUDA_COREDUMP: "1"
   SGLANG_JIT_DEEPGEMM_FAST_WARMUP: true
   SKIP_PR_TEST_HEALTH_CHECK: ${{ (fromJson(inputs.caller_inputs).skip_pr_test_health_check || fromJson(inputs.caller_inputs).test_parallel_dispatch || fromJson(inputs.caller_inputs).run_all_tests) && 'true' || 'false' }}

.github/workflows/nightly-72-gpu-gb200.yml

@@ -31,6 +31,7 @@ concurrency:
 
 env:
   SGLANG_IS_IN_CI: true
+  SGLANG_ENABLE_ASYNC_ASSERT: true
   SRT_SLURM_BRANCH: sglang-nightly-regression
   SLURM_PARTITION: batch
   SLURM_ACCOUNT: sglang

.github/workflows/nightly-test-nvidia.yml

@@ -46,6 +46,7 @@ concurrency:
 
 env:
   SGLANG_IS_IN_CI: true
+  SGLANG_ENABLE_ASYNC_ASSERT: true
   SGLANG_CUDA_COREDUMP: "1"
   HF_HUB_DOWNLOAD_TIMEOUT: 300
   HF_HUB_ETAG_TIMEOUT: 300

.github/workflows/rerun-test.yml

@@ -59,6 +59,7 @@ on:
 
 env:
   SGLANG_IS_IN_CI: true
+  SGLANG_ENABLE_ASYNC_ASSERT: true
   SGLANG_CUDA_COREDUMP: "1"
   SGLANG_JIT_DEEPGEMM_FAST_WARMUP: true
   # TEMP: rebuild deepep against the new torch for torch-211-merge PR only — revert before merging to main.

.github/workflows/weekly-test-nvidia.yml

@@ -20,6 +20,7 @@ concurrency:
 
 env:
   SGLANG_IS_IN_CI: true
+  SGLANG_ENABLE_ASYNC_ASSERT: true
   HF_HUB_DOWNLOAD_TIMEOUT: 300
   HF_HUB_ETAG_TIMEOUT: 300