Make gpu-specific build setup UX less bad

[magik6k]

Currently there is a whole lot of env variables that SPs need to set at build time to get things to build and work for their (usually nvidia) GPUs.

We could make the Makefile much easier to use by having some opinions:

• Build proofs from source by default (unless set to pull binaries which only really makes sense in CI
  • RUSTFLAGS="-C target-cpu=native -g" FFI_BUILD_FROM_SOURCE=1
• Default to SUPRASEAL C2 - it works fine, not using it when possible is objectively dumb
  • FFI_USE_CUDA_SUPRASEAL=1
  • May require also setting LIBRARY_PATH to point at local cuda install, arch-like it's /opt/cuda/targets/x86_64-linux/lib, ubuntu usually /usr/local/cuda/lib64
• Maybe detect old gpus (or have one simple flag for them) and set required vars
  • BELLMAN_CUDA_NVCC_ARGS="--fatbin --gpu-architecture=sm_61 --generate-code=arch=compute_61,code=sm_61"
  • NEPTUNE_CUDA_NVCC_ARGS="--fatbin --gpu-architecture=sm_61 --generate-code=arch=compute_61,code=sm_61"
  • EC_GPU_CUDA_NVCC_ARGS="--fatbin --gpu-architecture=sm_61 --generate-code=arch=compute_61,code=sm_61"
  • Old gpus can't use supraseal, so FFI_USE_CUDA=1