feat: add durable reconnect-safe experiment runs (#184)

tests/test_gpu_lifecycle.py

@@ -66,6 +66,43 @@ def test_disabled_publisher_is_noop():
     assert publisher.release() is None
 
 
+def test_allocation_phase_can_heartbeat_before_ready(tmp_path):
+    publisher = GpuLifecyclePublisher(tmp_path, run_id="loading-run")
+
+    def acknowledge():
+        while not (tmp_path / "current.json").exists():
+            time.sleep(0.001)
+        request = json.loads((tmp_path / "current.json").read_text())
+        (tmp_path / "ack.json").write_text(json.dumps({
+            "schema_version": 1,
+            "run_id": "loading-run",
+            "request_event_id": request["event_id"],
+            "decision": "grant",
+            "lease_id": "lease-loading",
+            "granted_vram_bytes": 100,
+        }))
+
+    thread = threading.Thread(target=acknowledge)
+    thread.start()
+    publisher.loading("org/model")
+    event = publisher.heartbeat()
+    publisher.release(reason="test_complete")
+    thread.join()
+
+    assert event is not None
+    assert event["event"] == "heartbeat"
+    assert event["run_id"] == "loading-run"
+
+
+def test_environment_uses_stable_experiment_run_id(tmp_path, monkeypatch):
+    from obliteratus.gpu_lifecycle import from_environment
+
+    monkeypatch.setenv("OBLITERATUS_GPU_LIFECYCLE_DIR", str(tmp_path))
+    monkeypatch.setenv("OBLITERATUS_RUN_ID", "run-stable")
+    publisher = from_environment()
+    assert publisher._run_id == "run-stable"
+
+
 def test_runtime_directory_must_exist(tmp_path):
     with pytest.raises(ValueError, match="must already exist"):
         GpuLifecyclePublisher(tmp_path / "missing")