Improve Vulkan dylib speed

### Prerequisites - [x] I am running the latest code. Mention the version if possible as well. - [x] I carefully followed the [README.md](https://github.com/Mozilla-Ocho/llamafile/blob/master/README.md). - [x] I searched using keywords relevant to my issue to make sure that I am creating a new issue that is not already open (or closed). - [x] I reviewed the [Discussions](https://github.com/Mozilla-Ocho/llamafile/discussions), and have a new and useful enhancement to share. ### Feature Description From the discussion [here](https://github.com/mozilla-ai/llamafile/pull/892#issuecomment-4147946006) it emerges that llamafile's vulkan seems to be about twice as slow as llama.cpp on the pp2048 test. There are different possible reasons for that, but one low hanging fruit could be to use better optimizations flags. ### Motivation Vulkan provides GPU support with a rather good portability, and compiles much faster than CUDA/ROCm. If we find a good build process for it and add it as a --rebuild option, this would allow people to have better acceleration with a rather low effort ### Possible Implementation _No response_