[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/spec_decode/eagle_proposer.py

@@ -123,7 +123,8 @@ class EagleProposer(Proposer):
                   num_reqs: int = 0,
                   num_tokens_across_dp: Optional[torch.Tensor] = None,
                   aclgraph_runtime_mode: CUDAGraphMode = CUDAGraphMode.NONE,
-                  batch_descriptor=None):
+                  batch_descriptor=None,
+                  dummy_compute_logits=lambda hidden_states: None):
         moe_comm_type = self.runner._select_moe_comm_method(num_tokens)
         with set_ascend_forward_context(None,
                                         self.vllm_config,
@@ -134,6 +135,7 @@ class EagleProposer(Proposer):
                 positions=self.positions[:num_tokens],
                 hidden_states=self.hidden_states[:num_tokens],
             )
+            dummy_compute_logits(self.hidden_states)
 
     def generate_token_ids(self,
                            valid_sampled_token_ids: list[np.ndarray],