bugfix for mtp (#3300)

### What this PR does / why we need it?
when mtp>1, we need refresh cos ans sin in each step.

### Does this PR introduce _any_ user-facing change?
no

### How was this patch tested?

- vLLM version: v0.11.0

Signed-off-by: zouyida2052 <zouyida2002@gmail.com>

vllm_ascend/attention/utils.py

@@ -97,7 +97,7 @@ def split_decodes_and_prefills(
         return num_reqs, 0, num_tokens, 0
 
     first_prefill = is_prefill.int().argmax(dim=-1).item()
-    assert torch.all(query_lens[first_prefill:] >= decode_threshold)
+    assert torch.all(query_lens[first_prefill:] > decode_threshold)
     assert torch.all(query_lens[:first_prefill] <= decode_threshold)
     num_decodes = first_prefill
     num_prefills = num_reqs - num_decodes