fix(warmup): fire vLLM warmup before planner so RunPod loads during planner+stones

fix(warmup): fire 1-token LLM ping at request start to warm RunPod GPU

fix(mellea): 120s cold-start timeout + drop run_reconcile fallback

fix(fsm): inject valid doc_ids into system prompt to prevent rag_npcc4 hallucination

fix(fsm): fallback uses non-streaming reconcile to avoid double hang

fix(fsm): fallback to non-strict reconcile when Mellea returns empty paragraph

fix(cornerstone): replace ThreadPoolExecutor with sequential loop — fixes Burr post-action cleanup hang

fix(burr): remove SQLitePersister cache — was poisoning state on first broken run

feat(burr): LocalTrackingClient, SQLitePersister cache, StepEventHook, conditional transitions

deploy(l4): self-contained Riprap mirror