[Core] Support pooling

[wangxiyuan]

This PR added pooling support for vllm-ascend

Tested with bge-base-en-v1.5 by encode:

from vllm import LLM

# Sample prompts.
prompts = [
  "Hello, my name is",
  "The president of the United States is",
  "The capital of France is",
  "The future of AI is",
]
# Create an LLM.
model = LLM(model="./bge-base-en-v1.5", enforce_eager=True)
# Generate embedding. The output is a list of EmbeddingRequestOutputs.
outputs = model.encode(prompts)
# Print the outputs.
for output in outputs:
    print(output.outputs.embedding)  # list of 4096 floats

Tested by embedding:

from vllm import LLM, SamplingParams

llm = LLM(model="./bge-base-en-v1.5", task="embed")
(output,) = llm.embed("Hello, my name is")

embeds = output.outputs.embedding
print(f"Embeddings: {embeds!r} (size={len(embeds)})")

Related: #200 #235

Known issue

The accuracy is not correct since this feature rely on enc-dec support. It'll be done in the following PR by @MengqingCao

[MengqingCao]

Test LLM.score with BAAI/bge-reranker-v2-m3 locally, and raised NotImplementedError:

[rank0]:   File "/home/xxx/code/vllm-cpu/vllm/vllm/attention/layer.py", line 220, in forward
[rank0]:     return self.impl.forward(self, query, key, value,
[rank0]:   File "/home/xxx/code/vllm-ascend/vllm_ascend/attention.py", line 546, in forward
[rank0]:     raise NotImplementedError("Encoder self-attention and "
[rank0]: NotImplementedError: Encoder self-attention and encoder/decoder cross-attention are not implemented for AscendAttentionBackendImpl

Seems we need to add the support of encoder self-attention and encoder/decoder cross-attention.
But I'm fine with this pr for initial pooling model support, we can do the rest of the work in the following PRs

[wangxiyuan]

Test LLM.score with BAAI/bge-reranker-v2-m3 locally, and raised NotImplementedError:

[rank0]:   File "/home/xxx/code/vllm-cpu/vllm/vllm/attention/layer.py", line 220, in forward
[rank0]:     return self.impl.forward(self, query, key, value,
[rank0]:   File "/home/xxx/code/vllm-ascend/vllm_ascend/attention.py", line 546, in forward
[rank0]:     raise NotImplementedError("Encoder self-attention and "
[rank0]: NotImplementedError: Encoder self-attention and encoder/decoder cross-attention are not implemented for AscendAttentionBackendImpl

Seems we need to add the support of encoder self-attention and encoder/decoder cross-attention. But I'm fine with this pr for initial pooling model support, we can do the rest of the work in the following PRs

BAAI/bge-reranker-v2-m3 rely on Encode-only attention which is not support yet. I think we can do it in enc-dec feature.