Removed support for new NVIDIA graphics cards by downgrading CUDA version.

[LeXaMeN]

I installed your project a couple of months ago. I somehow figured out how everything works. I decided to install it again now...

If I were doing this from scratch now, I'd be cursing everyone, "Why isn't it working?"

4e455d9#diff-0fa8f631eeec4a7ed71628660621488bb1de7aa4c5d9906149bd6f900b443cac

You also removed the default config passthrough. Newbies will definitely go crazy after building Docker. Why do they change config.json, rebuild, and nothing changes?

• You can create a new profile like gpu50 (50XX) or gpu128 (CUDA 128) to support new NVIDIA cards.

[LeXaMeN]

Also create bug about build cuda 12.8 cards 50XX

ggml-org/llama.cpp#18430