Provide an option to prevent sending the system_prompt to the LLM, allowing the server to use its default system prompt, which can prevent unexpected behavior and context size issues.
### Feature request Make it possible to not send system_prompt ### Motivation when I started using avante, I was surprised to see "weird" answers from the LLM, at least different from the ones used in the various providers web chat UIs. When I started using a local instance of llama.cpp, my first message "hello world" returned a message "exceeding context size" (of 8k at the time) and that's when I realized the prompt was opiniated and sent tons of stuff (there are several issues about this, including https://github.com/yetone/avante.nvim/issues/2421). When using llama-server, in absence of system_prompt, it loads the one defined by the model which looks ideal for my usecase. for instance if you take qwen 2.5 - 7b, the default template checks for "system" role and has a fallback: ``` {%- if messages[0]['role'] == 'system' %}\n {{- messages[0]['content'] }}\n {%- else %}\n {{- 'You are Qwen, created by Alibaba Cloud. Y https://github.com/yetone/avante.nvim/issues?q