Centralize cuda_arch capability definition

[Copilot]

Wait for:

• feat: build arrow with cuda for eic_cuda #199

cuda_arch=75 was duplicated across 5 package specs in cuda/spack.yaml and tf/spack.yaml, making it hard to retarget GPU architecture.

Changes

• New spack-environment/cuda_arch.yaml: Single source of truth for the CUDA architecture target (currently cuda_arch=75, Compute Capability 7.5 / Turing). Change GPU target here only.
• cuda/spack.yaml and tf/spack.yaml: Include ../cuda_arch.yaml; drop inline cuda_arch=75 from acts, arrow, celeritas, py-torch, and py-tensorflow specs.

# spack-environment/cuda_arch.yaml
# Current target: Compute Capability 7.5 (Turing: T4, RTX 2xxx, Quadro RTX)
packages:
  all:
    variants: cuda_arch=75

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.

[wdconinc]

Looks sensible. Maybe there's a risk this will enable cuda_arch and therefore +cuda for packages where we currently don't explicitly enable +cuda, but would that be a bad thing?

[Copilot]

Pull request overview

This PR centralizes the CUDA compute capability setting for the Spack CUDA-related environments by introducing a shared cuda_arch.yaml include and removing per-spec cuda_arch=75 pins.

Changes:

• Add a new spack-environment/cuda_arch.yaml and include it from CUDA/TensorFlow environments.
• Remove explicit cuda_arch=75 from CUDA-enabled specs in spack-environment/cuda/spack.yaml.
• Remove explicit cuda_arch=75 from the TensorFlow CUDA spec in spack-environment/tf/spack.yaml.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
spack-environment/tf/spack.yaml	Includes centralized CUDA arch config; drops per-spec cuda_arch=75 on TensorFlow.
spack-environment/cuda/spack.yaml	Includes centralized CUDA arch config; drops per-spec cuda_arch=75 on multiple CUDA specs.
spack-environment/cuda_arch.yaml	New shared config intended to define CUDA architecture in one place.

---

You can also share your feedback on Copilot code review. Take the survey.

[Copilot]

packages: all: variants: cuda_arch=75 will try to apply the cuda_arch variant to every package in the environment, including many CPU-only packages that don’t define cuda_arch, which can break concretization. This also conflicts with the established repo pattern in spack-environment/packages.yaml (see the comment about avoiding packages:all:variants and using require:any_of to avoid unsupported variants). Consider constraining the setting to CUDA-enabled specs only (e.g., via a conditional require with when: '+cuda', or by keeping cuda_arch=75 on the specific CUDA specs / per-package requires).

Suggested change

	variants: cuda_arch=75
	require:
	- when: '+cuda'
	any_of: [cuda_arch=75, '@:']