fix: fix incorrect attribute for CPU sequence lengths

I'm just so tired of fixing these.

Signed-off-by: Yizhou Liu <[email protected]>

Author: yiz-liu

vllm_ascend/spec_decode/mtp_proposer.py

@@ -1107,7 +1107,7 @@ def prepare_inputs_padded(
         spec_common_attn_metadata = AscendCommonAttentionMetadata(
             query_start_loc=common_attn_metadata.query_start_loc,
             query_start_loc_cpu=query_start_loc_cpu,
-            seq_lens_cpu=common_attn_metadata.seq_lens,
+            seq_lens_cpu=common_attn_metadata.seq_lens_cpu,
             num_reqs=common_attn_metadata.num_reqs,
             num_actual_tokens=total_num_tokens,
             num_input_tokens=common_attn_metadata.num_input_tokens,