Add support for image input for multimodal Models

Images can not be simply sent through prompts, there should be field in additonal properties for text generator or think about a new node that takes image/url as a paramer and we can image URL or its base64 encoding as an input to that.