harness2: per-role models, sweep batch, task file assets endpoint

- Role models: run_improvement_via_tasks gains role_models
  (probe/editor/selector/reflector) plus --model-<role> CLI flags;
  missing/empty falls back to shared --model, both empty means
  server default, so default behavior is unchanged.
- Sweep: run_sweep + examples/harness2_sweep.py manifest CLI with
  shared library_root (library restore), per-run failure isolation
  and sweep-report.json summary; preflight gates task creation.
- Deliverable assets: GET /tasks/{id}/files for local-executor
  tasks (safe path resolution, UTF-8 text <=256KB, listing <=200;
  docker/cloud/cross-host 409 with explicit reason, since tenant
  Docker shares one /app/work per project and per-task attribution
  would misattribute); probe rollout deliverable_files from the
  endpoint plus DELIVERABLE FILE reply blocks (endpoint wins on
  collision, bad blocks dropped), surfaced in
  deliverables/<label>/files/ context and full_result summary.
- Docs: harness2.md comparison table updated for all three items.

examples/harness2_run.py

@@ -69,6 +69,10 @@ def main():
     ap.add_argument("--output", default="")
     ap.add_argument("--library-root", default="")
     ap.add_argument("--model", default=None)
+    ap.add_argument("--model-probe", default=None)
+    ap.add_argument("--model-editor", default=None)
+    ap.add_argument("--model-selector", default=None)
+    ap.add_argument("--model-reflector", default=None)
     ap.add_argument("--effort", default=None)
     ap.add_argument("--approval", default=None)
     ap.add_argument("--environment-id", default=None)
@@ -148,7 +152,11 @@ def main():
         k=args.k, mode=args.mode,
         library_root=Path(args.library_root) if args.library_root else None,
         budget_usd=args.budget, max_sessions=args.max_sessions,
-        model=args.model, effort=args.effort,
+        model=args.model,
+        role_models={"probe": args.model_probe, "editor": args.model_editor,
+                     "selector": args.model_selector,
+                     "reflector": args.model_reflector},
+        effort=args.effort,
         approval=args.approval, environment_id=args.environment_id,
         executor=args.executor, no_time_limit=args.no_time_limit,
         timeout_seconds=args.timeout)