Specify img2text model by tag

[hrchu]

What problem does this PR solve?

The current design is not well-suited for multimodal models, as each model can only be configured for a single purpose—either chat or Img2txt. To work around this limitation, we use model aliases such as gpt-4o-mini and gpt-4o-mini-2024-07-18.

To fix this, this PR allows specifying the Img2txt model by tag instead of model_type.

Type of change

• [x] Refactoring