Support InternVL2 Series

[amosyou]

Motivation

This PR adds support for InternVL2 series of models. Currently working on InternVL2-8B.

Modifications

These are the main modifications I'm making from the vllm implementation.

• new chat template
• add dummy image processor because of dynamic processing
• patch type -> rope_type in rope_scaling due to vllm standardization

Some broader design questions I have as I'm implementing this.

1. Due to the dynamic processing feature of InternVL2, there is no image_processor that we load. Since these transforms are done on the fly with respect to each individual image, I'm using a dummy image processor, and transforming the images in the forward pass of the model. This should be fine right?
2. Do we want to remove vllm rope dependency now or can that be done in a later commit?

TODOs

• pad_input_ids in forward

Checklist

• Format your code according to the Contributor Guide.
• Add unit tests as outlined in the Contributor Guide.
• Update documentation as needed, including docstrings or example tutorials.