[Feat] Dynamic Batch Feature (#3490)

[RFC](https://github.com/vllm-project/vllm-ascend/issues/3328) for more
details.
Add dynamic batch feature in chunked prefilling strategy, the token
budget can be refined to achieve better effective throughput and TPOT.

!!! NOTE: only 910B3 is supported till now, we are working on further
improvements.
Additional file for lookup table is required.

- vLLM version: v0.11.0rc3
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0

---------

Signed-off-by: Cheng Wang <wangchengkyrie@outlook.com>

vllm_ascend/ascend_config.py

@@ -129,6 +129,8 @@ class AscendConfig:
             if self.pd_tp_ratio == 0:
                 raise AssertionError(
                     "Only support P node tp size lagger then D node tp size")
+        self.SLO_limits_for_dynamic_batch = additional_config.get(
+            "SLO_limits_for_dynamic_batch", -1)
 
 
 class TorchairGraphConfig: