DocVQA inference support

[ghost]

This pr integrates (DocVQA) functionality into the Florence2 ComfyUI node.

Key Features:

• Downloads the DocVQA Florence 2 fine-tune from HF with the existing Florence2 node
• Adds DocVQA as a new task option within the existing Florence2 node
• Enables users to perform document-based visual question answering using the Florence2 model
• Expands the versatility of the Florence2 node in ComfyUI

Implementation Details:

• Integrated DocVQA inference as a new task type for the Florence2 model
• Added necessary preprocessing for document images
• Implemented post-processing to format model outputs into human-readable answers
• Updated the model selection to include DocVQA-capable Florence2 variants

Usage: Users can now select DocVQA as a task when using the Florence2 node in ComfyUI. They can input document images, specify their questions, and receive AI-generated answers based on the document's content.

Example:

[kijai]

Nice, I was unaware of this one. Thank you!