Update to latest llama.cpp

[cebtenzzre]

NOTE: Requires this llama.cpp branch, which is where most of the changes are: https://github.com/nomic-ai/llama.cpp/tree/update-llamacpp-base
We will have to reset our llama.cpp fork's master branch to that commit before we can properly merge this.

What I did:

• Split the Vulkan changes into their own branch: https://github.com/nomic-ai/llama.cpp/tree/vulkan
• Incrementally merged the latest llama.cpp into them (up to ggerganov/llama.cpp@6b0a7420), making commits as needed to maintain parity between Metal and Vulkan. This is the current state of the vulkan branch.
• Merged the vulkan branch into the GPT4All-specific branch, and made several changes on top of that. This is update-llamacpp-base.
• Updated the llama.cpp dependency in GPT4All and made the necessary changes for it to be used. That is this PR.

It seems to run fine with app.py on my RX 7800 XT. 0cc4m tested a slightly older version of the changes to llama.cpp and also had success (after some initial failures that we couldn't reproduce).