fix(model): predict method for url/base64

[Phelan164]

Because

• there is an endpoint for inference with URL or base64

This commit

• update the REST endpoint and message request

[bochengyang]

@Phelan164 please refer to pipeline-backend's proto file, users then can have a unified request for file/URL/base64 inference/predict

1. The content of the request
2. The inference/predict request

[Phelan164]

@bochengyang I checked on pipeline-backend. However, there is one issue when unifying all message for both stream method and rest api.

• bytes content is required for stream gRPC method but url/base64 is required for normal method.
  So from unified struct, it do not show which message field is required for each method

[xiaofei-du]

@bochengyang I checked on pipeline-backend. However, there is one issue when unifying all message for both stream method and rest api.

• bytes content is required for stream gRPC method but url/base64 is required for normal method.
  So from unified struct, it do not show which message field is required for each method

I think it makes sense to separate messages for stream and rest API. It won't confuse the user.

[xiaofei-du]

LGTM 👍