fix: simplify GMS layout state and harden GPU-backed flows (#7006)

Signed-off-by: mohammedabdulwahhab <furkhan324@berkeley.edu>
Signed-off-by: Hannah Zhang <hannahz@nvidia.com>
Co-authored-by: mohammedabdulwahhab <furkhan324@berkeley.edu>
Co-authored-by: hhzhang16 <54051230+hhzhang16@users.noreply.github.com>

tests/report_pytest_markers.py

@@ -120,10 +120,35 @@ STUB_MODULES = [
     "botocore.exceptions",
     "pynvml",
     "gpu_memory_service",
+    "gpu_memory_service.client",
+    "gpu_memory_service.client.memory_manager",
+    "gpu_memory_service.client.rpc",
+    "gpu_memory_service.client.session",
+    "gpu_memory_service.client.torch",
+    "gpu_memory_service.client.torch.allocator",
+    "gpu_memory_service.client.torch.module",
+    "gpu_memory_service.client.torch.tensor",
     "gpu_memory_service.common",
+    "gpu_memory_service.common.cuda_utils",
+    "gpu_memory_service.common.protocol",
+    "gpu_memory_service.common.protocol.messages",
+    "gpu_memory_service.common.protocol.wire",
+    "gpu_memory_service.common.types",
     "gpu_memory_service.common.utils",
     "gpu_memory_service.failover_lock",
     "gpu_memory_service.failover_lock.flock",
+    "gpu_memory_service.integrations",
+    "gpu_memory_service.integrations.common",
+    "gpu_memory_service.integrations.common.utils",
+    "gpu_memory_service.integrations.sglang",
+    "gpu_memory_service.integrations.sglang.memory_saver",
+    "gpu_memory_service.integrations.vllm",
+    "gpu_memory_service.integrations.vllm.worker",
+    "gpu_memory_service.server",
+    "gpu_memory_service.server.allocations",
+    "gpu_memory_service.server.gms",
+    "gpu_memory_service.server.rpc",
+    "gpu_memory_service.server.session",
     "prometheus_client",
     "prometheus_client.parser",
     "sklearn",