test: give the logprobs scheduler doubles the vLLM 0.26 contracts

test_omni_ar_scheduler_logprobs was missed by the 0.26 mock sweep
(bc46e54a / 763aa865): its _Request double lacks num_in_flight_tokens
(read by the update_from_output settling loop) and its _free_request
stub returns None instead of the (kv_xfer_params, ec_xfer_params)
2-tuple, so test_invalid_logprobs_finish_only_the_affected_scheduler_request
fails on any 0.26-based tree. Reproduces on origin/main.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
(cherry picked from commit 05d0926e6d20edc026906b96b561443a0d669904)

tests/core/sched/test_omni_ar_scheduler_logprobs.py

@@ -76,6 +76,7 @@ class _Request:
         self.status = RequestStatus.RUNNING
         self.sampling_params = SimpleNamespace(num_logprobs=1)
         self.num_computed_tokens = 0
+        self.num_in_flight_tokens = 0
         self.num_output_placeholders = 0
         self.has_encoder_inputs = False
         self.pooling_params = None
@@ -137,7 +138,8 @@ def test_invalid_logprobs_finish_only_the_affected_scheduler_request() -> None:
         scheduler.requests.pop(request.request_id, None)
         scheduler.finished_req_ids.add(request.request_id)
         scheduler.finished_req_ids_dict[request.client_index].add(request.request_id)
-        return None
+        # vLLM 0.26 contract: (kv_xfer_params, ec_xfer_params)
+        return None, None
 
     scheduler._update_request_with_output = update_request
     scheduler._process_kv_transfer_trigger = lambda _request, _tokens: False