Run loader diagnosis as background job

app.py

@@ -420,6 +420,10 @@ def _run_benchmark_job(job_id, kind, model_filter=None, force=False):
             with _BENCH_LOCK:
                 result = _run_cpu_smoke(model_filter)
             payload = {"status": "complete", "result": result}
+        elif kind == "diagnose":
+            with _BENCH_LOCK:
+                result = _run_cpu_diagnose(model_filter)
+            payload = {"status": "complete", "result": result}
         else:
             if force:
                 global _BENCH_RESULTS
@@ -519,12 +523,12 @@ def _patched_create_app(blocks, **kwargs):
         except Exception as e: return JSONResponse({"error": str(e)}, status_code=502)
 
     @fa_app.get("/respite/benchmark/diagnose")
-    def _diagnose(model: str="TinyLlama 1.1B (control)"):
-        try:
-            with _BENCH_LOCK:
-                return JSONResponse(_run_cpu_diagnose(model.strip()))
-        except Exception as e:
-            return JSONResponse({"error": str(e)}, status_code=500)
+    def _diagnose(model: str="TinyLlama 1.1B (control)", job_id: str=""):
+        if job_id:
+            with _BENCH_JOBS_LOCK:
+                return JSONResponse(_BENCH_JOBS.get(job_id, {"status": "not_found"}))
+        job = _start_benchmark_job("diagnose", model.strip())
+        return JSONResponse({"status": "accepted", "job_id": job, "poll": "/respite/benchmark/diagnose?job_id=" + job}, status_code=202)
 
     @fa_app.get("/respite/benchmark/smoke")
     def _smoke(model: str="Gemma 4 E4B", job_id: str=""):