Drop CUDA 11.8

CUDA 11.8 is seeing less and less use (not least because key feedstocks like pytorch, tensorflow, jax, onnx have dropped it already). Moreover, it's the last hold-over from before the shiny new [infrastructure](https://github.com/conda-forge/conda-forge.github.io/issues/1963) we got with CUDA 12. Dropping 11.8 came up recently in #6917 already; one concern that was brought up is > @jakirkham: We do this in part because [CUDA 11.8 is the last version that still supports Kepler](https://docs.nvidia.com/cuda/archive/11.8.0/cuda-toolkit-release-notes/index.html#title-new-cuda-libraries). In [CUDA 12.0, Kepler support was removed](https://docs.nvidia.com/cuda/archive/12.0.0/cuda-toolkit-release-notes/index.html#cuda-libraries). Once we drop CUDA 11.8, those users will no longer have support I don't know how long they were being sold, but Kepler GPUs are from 2012 ([wiki](https://en.wikipedia.org/wiki/Kepler_(microarchitecture))); while dropping compat is never great, I think this is an acceptably long time for dropping old HW support. Dropping 11.8 would simplify our pinning quite substantially (c.f. #6967), because we won't need to zip the images into the compilers anymore. It would also unblock migrations like #7005, which is the most immediate motivation for doing this sooner rather than later (but the other benefits are not to be sneezed at). A comment from that PR: > @jakirkham: Before dropping CUDA 11.8 we should try to assess the amount of usage it still sees somehow by maintainers and users. If we confirm this to be low, then document these findings and add a news item. Xref: #6980 Xref: #5339 CC @conda-forge/cuda @hmaarrfk