Vision Service: local processer for labels

[hardyliao85]

Summary
Enable label generation on PhotoPrism-Vision to offload inference from the main service and allow the use of more advanced classifier models.

Describe what problem this solves and why this would be valuable to many users
Currently, PhotoPrism can only perform image classification inference on the main service and cannot generate labels through PhotoPrism-Vision. If label generation could be handled by PhotoPrism-Vision, users would be able to leverage more powerful classifier models for more accurate results.

Describe the solution you'd like
The goals include the following:

1. Similar to how captions are configured in vision.yml, users should be able to specify the correct name and call API /api/v1/vision/labels to receive label results.
2. Allow the language of labels to be specified in Docker Compose.

Describe alternatives you've considered
Introduce different levels of classifier models so users can choose the appropriate model based on their hardware capabilities.

Additional context
I am very interested in PhotoPrism, but since the main service is running on a device with relatively low hardware specifications, I need to offload vision processing to other devices. I am currently experimenting with developing this feature and hope that PhotoPrism-Vision can become more complete.
I don’t have much experience with submitting Git PRs, so if there’s anything missing or unclear in this request, please feel free to let me know. Thank you!