FEAT: support sparse vector for bge-m3

[pengjunfeng11]

支持bge-m3模型的稀疏向量生成功能，

调用方式为：model.create_embedding(text, return_sparse=True)

新增convert_ids_to_tokens方法

该方法可将token_id转换为人类可读文字，调用方式为

from xinference.client import Client

client = Client("http://ip:port")
model = client.get_model(model_name)
seq = model.convert_ids_to_tokens(key_list)

该方法返回类型为List[str]，如传入List[str]，将按顺序返回值

Fixes #2527 .

[qinxuye]

For convert_ids_to_tokens, can you add a test to verify it to

https://github.com/xorbitsai/inference/blob/main/xinference/model/embedding/tests/test_embedding_models.py

For bge-m3, it's too large for CI, we can test it manually.

[qinxuye]

This looks a bit disruptive to the design, I don't know if there is a more elegant way.

[pengjunfeng11]

Are you referring only to the if judgment part or the subsequent reload part?

[qinxuye]

I mean the reload part.

[qinxuye]

How about loading bge-m3 when specifying hybrid_mode=True? This can be done in load.

[pengjunfeng11]

For convert_ids_to_tokens, can you add a test to verify it to

main/xinference/model/embedding/tests/test_embedding_models.py

For bge-m3, it's too large for CI, we can test it manually.

OK