Dropping cuda-cudart_{{ target_platform }} from requirements/run

[jakirkham]

Recently cuda-cudart_{{ target_platform }} was added as a workaround to an overlinking (or really underlinking) error with cuda-cudart on Windows ( conda-forge/cuda-cudart-feedstock#13 )

Since then the upstream issue was fixed, so it should be possible to revert this workaround. Namely dropping this line from the recipe/meta.yaml:

llama.cpp-feedstock/recipe/meta.yaml

Line 75 in c4127c5

- cuda-cudart_{{ target_platform }} {{ cuda_compiler_version }} # [(cuda_compiler_version or "").startswith("12") and win]