Solved the bug of inference failure when target_path is set but there is no model in it

[JingofXin]

For models that are only deployed, lazy deployment is used on a single machine, and there is a merge folder in the current path, merge will be used first. This will cause errors in model loading.

【Bug】
Reproduction conditions:

• Single machine
• There is a merge folder in the current path
• Only deploy models
  

【Solve】

[wzh1994]

如线下沟通，这里的逻辑统一改成：

1. 当设置了base-model和target-path时，优先部署base-model
2. 用户可以指定用target-path中的特定微调版本部署
3. 当调用update进行微调时，微调结束立即设置当前版本为部署版本，然后进行部署和评测

此外应注意每个模型可以起1-多个微调任务，多个任务相互不影响