+cuda cuda_arch==XY does not properly propagate cuda_arch in some cases

Steps to reproduce

Using:

• spack@develop (27f378601eeb8472f40379b83e28989e1ca260c6 from Mon Jul 31 09:54:35 2023 -0500)
• Ubuntu 20.04, X86_64

$> spack config add "concretizer:reuse:false"
$> spack spec -I py-cupy +cuda cuda_arch==80
==> Error: concretization failed for the following reasons:

   1. NCCL requires CUDA

It works if I manually specify:

$> spack spec -I py-cupy +cuda cuda_arch=80 ^nccl+cuda cuda_arch=80
OK

Error message

No response

Information on your system

• Spack: 0.21.0.dev0 (27f378601eeb8472f40379b83e28989e1ca260c6)
• Python: 3.8.10
• Platform: linux-ubuntu20.04-cascadelake
• Concretizer: clingo

General information

FYI @becker33 @alalazo

• I have run spack debug report and reported the version of Spack/Python/Platform
• I have searched the issues of this repo and believe this is not a duplicate
• I have run the failing commands in debug mode and reported the output

Reproduce the failure with spack spec -I py-cupy +cuda cuda_arch==80, then compare it with the working command that explicitly sets ^nccl+cuda cuda_arch=80. Trace Spack's concretization and CUDA architecture propagation for py-cupy and nccl; done means the original command concretizes successfully without manually specifying NCCL's CUDA settings.