Docker improvements

[Isydmr]

• Add --gpus:all parameter to CuBLAS to avoid The NVIDIA Driver was not detected. GPU functionality will not be available error while running cuda_simple
• Move CuBLAS Nvidia support note under cuda_simple
• Use Markdown extensions for notes (important, note, etc.)
• Remove USE sections at the end
• Readability improvements (use third-level heading mostly)

Rendered version: https://github.com/Isydmr/llama-cpp-python/tree/main/docker

[Isydmr]

@abetlen ready to review/merge

[abetlen]

@Isydmr thank you, looks good!