[Bugfix] Resolve MTP > 1 issue when lm head tp > 1 (#4254)

### What this PR does / why we need it?

Previously, the dummy run executed compute_logits only once, regardless
of num_speculative_tokens. This caused execute_model to hang on
compute_logits when lm head tensor parallelism exceeded 1. The fix
ensures compute_logits executes correctly during dummy run, matching
num_speculative_tokens.

I set the `non_blocking` argument to False when moving
`exceeds_max_model_len` to the CPU. From what I understand, using
`non_blocking=True` and immediately accessing the tensor on the CPU can
cause accuracy problems. However, this issue doesn't happen when
transferring data to a device. ref:
https://discuss.pytorch.org/t/should-we-set-non-blocking-to-true/38234/18

- vLLM version: v0.11.0
- vLLM main:
2918c1b49c

---------

Signed-off-by: Jade Zheng <zheng.shoujian@outlook.com>

vllm_ascend/torchair/torchair_mtp_proposer.py

@@ -81,7 +81,8 @@ class TorchairMtpProposer(MtpProposer):
                   num_reqs: int = 0,
                   num_tokens_across_dp=None,
                   aclgraph_runtime_mode: CUDAGraphMode = CUDAGraphMode.NONE,
-                  batch_descriptor=None) -> None:
+                  batch_descriptor=None,
+                  dummy_compute_logits=lambda hidden_states: None) -> None:
         moe_comm_type = self.runner._select_moe_comm_method(num_tokens)
 
         if not with_prefill:
@@ -143,6 +144,7 @@ class TorchairMtpProposer(MtpProposer):
                     self.model(input_ids=input_ids,
                                positions=positions,
                                hidden_states=previous_hidden_states)
+                dummy_compute_logits(previous_hidden_states)
             if with_prefill:
                 break