[UI]: Support in-server model proxy

[jvstme]

Context

#1595 adds support for running models without a gateway, in which case the model is served by dstack-server and the model URL consists only of a dstack-server path. Frontend changes are required to correctly display such URLs on the Models page (/models) and use them with the OpenAI client in the model chat.

Current behavior

• The table on the Models page has a column named "Gateway".
• The service.model.base_url field returned by /api/runs/get or /api/runs/list always contains a full URL that is then displayed in the "Gateway" column and passed to the OpenAI client as is.

New expected behavior

• The column should be renamed "URL".
• The service.model.base_url field may contain either a full URL or only a path.
  • If it's a full URL, it should be displayed in the "URL" column and passed to the OpenAI client as is.
  • If it's a path, the URL displayed in the "URL" column and passed to the OpenAI client should be constructed by adding that path to the dstack-server base URL. ¹

Examples

dstack-server base URL	The service.model.base_url field	Resulting model URL
any	https://gateway.admin.sky.dstack.ai	https://gateway.admin.sky.dstack.ai
any	http://gateway.mycompany.example	http://gateway.mycompany.example
https://sky.dstack.ai	/proxy/models/admin/	https://sky.dstack.ai/proxy/models/admin/
http://localhost:3000	/proxy/models/admin/	http://localhost:3000/proxy/models/admin/

Footnotes

1. The dstack-server base URL here means the scheme, hostname, and optional port. The /api prefix is not part of the base URL. ↩