Add context window size information to client capabilities

[ekzhu]

What feature would you like to be added?

Right now, model capabilities include:

        ModelCapabilities(
            function_calling=True, json_output=True, vision=True
        )

We should add other information that might be needed upstream such as the context window size, the max number of output tokens, perhaps the cost (input and output), as well as the max number of images, the max size of images, etc.

Why is this needed?

This metadata allows application to make decision about context input size, and other parts of the framework to manage context.