[Refactor] remove some metadata variables in attention_v1. (#5160)

RFC: https://github.com/vllm-project/vllm-ascend/issues/4629

Reason:

The metadata data class contains an excessive number of variables. We
will inherit the metadata of the community and simultaneously remove
some variables that are no longer needed at present.

Todo:
1. remove attn_state partly.

- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c

---------

Signed-off-by: weijinqian_v1 <weijinqian@huawei.com>
Co-authored-by: weijinqian_v1 <weijinqian@huawei.com>

vllm_ascend/compilation/acl_graph.py

@@ -271,8 +271,8 @@ def _update_attn_fia_params(update_stream, forward_context, runtime_shape):
              attn_output, softmax_lse) = param
 
             seq_lens = forward_context.attn_metadata[key].seq_lens_list
-            query_start_loc = forward_context.attn_metadata[
-                key].query_start_loc_list
+            actual_seq_lengths_q = forward_context.attn_metadata[
+                key].actual_seq_lengths_q
             torch.npu.graph_task_update_begin(update_stream, handle)
             torch_npu.npu_fused_infer_attention_score.out(
                 query=query,
@@ -282,7 +282,7 @@ def _update_attn_fia_params(update_stream, forward_context, runtime_shape):
                 atten_mask=attn_mask,
                 input_layout="TND",
                 block_size=block_size,
-                actual_seq_lengths=query_start_loc,
+                actual_seq_lengths=actual_seq_lengths_q,
                 actual_seq_lengths_kv=seq_lens,
                 num_key_value_heads=num_kv_heads,
                 num_heads=num_heads,