Update granite vision docs for 3.2 model

[alex-jw-brooks]

This PR updates the granite vision docs with a few changes, outlined below:

• Updates docs from the 3.1 preview -> the newly released 3.2 2b model!
• Updates the image grid pinpoints - there is a small change in the image grid pinpoints from the 3.1 preview
• Adds guidance on quantization; the LLM can be quantized, but the visual encoder can't be since siglip has tensor dims nondivisible by 32, which could be a point of confusion for some, so adding some explicit guidance
• Changes the example image to use the llama cpp banner, as the current link is no longer valid, and this model is good for things like doc understanding anyway

@danbev can you please take a look when you have a moment? 🙂