feat: Prompt and LLM should accept images

[ludwiktrammer]

• The BasePrompt class should include a new method for returning a list of images (default implementation: empty list). It's important that the format in which the images is compatible with how images are stored in Document Search's ImageDocument (although Prompt shouldn't directly depend on ImageDocument)
• The Prompt class should include its own implementation of the method, that automatically returns images from chosen fields of the Prompt's input model (which fields should probably be configured as a static field in the Prompt class)
• The LLM and LLMClient implementations (including the built-in implementation using LiteLLM) should automatically send those images to LLM models that support it (see: the Vision section in OpenAI docs and Checking if a model supports vision in LiteLLM docs