[BodhiApp] changes for bodhiapp.

[21-dec-24] using prompt if passed in chat completions, and not using messages
- added python integration tests for changes in server
- having add_special as request param to pass to upstream, allows pre-formatted chat messages to not be formatted again
- modified workflow to download and cache the llama2-7b model used for integration testing
[17-jan-25] updated to latest llama.cpp
server.cpp had changes where ctx_server.vocab was used instead of the earlier ctx_server.ctx
[26-jan-25] merged examples/server/utils.hpp, llama.cpp is starting to support tools in the chat template, resolved rebase conflict by merging our logic of preformatted messages with new feature
adding the github workflow to create the artifacts based on platform and gpu architecture, for macos-cpu+metal, ubuntu-cpu+cuda, windows-cpu+cuda-12.4+11.6
also added for other ubuntu+windows cpu variants and gpu archs like vulkan etc., but disabled for now