Optimized DeepSeek-V2 attention prefill with MHA.

[gyou2021]

The MHA without cache compression replaces MLA for the prefill stage of DeepSeek-V2, significantly reducing computation costs, especially for long sequence inputs during inference. The time to the first token can be reduced.

command example：
python run_generation.py --model_name_or_path --bf16 --trim_logits --batch_size 4
--max_input_tokens 3072 --max_new_tokens 256 --use_hpu_graphs --use_kv_cache --reuse_cache
--bucket_size 128 --bucket_internal --limit_hpu_graphs

What does this PR do?

Fixes # (issue)

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?