Adds support for only the llama3.2 vision models on groq. This comes with many conditionals and nuances to handle as Groqs vision implemention is quite bad right now
