[ModelRunner] remove unused args (follow vllm changes) (#159)

### What this PR does / why we need it?
The arg list of `Attention.forward()` is changed by
vllm-project/vllm#13555.
The unused args `kv_caches` and `attn_metadata` are removed.

### Does this PR introduce _any_ user-facing change?
N/A

### How was this patch tested?
CI passed with existing test.

Signed-off-by: MengqingCao <[email protected]>

Author: MengqingCao

vllm_ascend/model_runner.py

@@ -1142,8 +1142,6 @@ def execute_model(
                 hidden_or_intermediate_states = model_executable(
                     input_ids=model_input.input_tokens,
                     positions=model_input.input_positions,
-                    kv_caches=kv_caches,
-                    attn_metadata=model_input.attn_metadata,
                     intermediate_tensors=intermediate_tensors,
                     **MultiModalKwargs.as_kwargs(multi_modal_kwargs,
                                                  device=self.device),