[CI/Build] Keep non-CUDA wheels CUDA-free

[Aionw]

Description

The published 0.3.13 non-CUDA wheel still links mooncake_client and
store.so against libcudart.so.12. Non-CUDA wheels use a CUDA manylinux
image as their toolchain, and Store was treating the presence of that toolkit
as permission to enable CUDA staging.

This change:

• enables Store CUDA/HIP staging only when the corresponding project option is
  selected;
• requires the selected CUDA/HIP toolkit to be available;
• scans every Mooncake artifact in an installed non-CUDA wheel for CUDA runtime
  dependencies instead of checking only mooncake_master.

Module

• Transfer Engine (mooncake-transfer-engine)
• Mooncake Store (mooncake-store)
• Reshard (mooncake-reshard)
• Mooncake EP (mooncake-ep)
• Mooncake PG (mooncake-pg)
• Integration (mooncake-integration)
• P2P Store (mooncake-p2p-store)
• Python Wheel (mooncake-wheel)
• Common (mooncake-common)
• Mooncake RL (mooncake-rl)
• CI/CD
• Docs
• Other

Type of Change

• Bug fix
• New feature
• Refactor
• Breaking change
• Documentation update
• Performance improvement
• Other

How Has This Been Tested?

Test commands:

cmake -S . -B /tmp/mooncake-noncuda-config -G Ninja \
  -DBUILD_UNIT_TESTS=OFF -DBUILD_BENCHMARK=OFF \
  -DUSE_CUDA=OFF -DUSE_HTTP=ON -DUSE_ETCD=OFF \
  -DSTORE_USE_ETCD=OFF -DWITH_EP=OFF -DWITH_STORE_RUST=OFF
cmake --build /tmp/mooncake-noncuda-config --parallel 128 \
  --target mooncake_client store
readelf -d /tmp/mooncake-noncuda-config/mooncake-store/src/mooncake_client
readelf -d /tmp/mooncake-noncuda-config/mooncake-integration/store.cpython-310-x86_64-linux-gnu.so

cmake -S . -B /tmp/mooncake-cuda-config -G Ninja \
  -DBUILD_UNIT_TESTS=OFF -DBUILD_BENCHMARK=OFF \
  -DUSE_CUDA=ON -DUSE_HTTP=ON -DUSE_ETCD=OFF \
  -DSTORE_USE_ETCD=OFF -DWITH_EP=OFF -DWITH_STORE_RUST=OFF

pre-commit run --files \
  .github/workflows/_build-wheel.yaml \
  mooncake-store/src/CMakeLists.txt

Test results:

• Unit tests pass
• Integration tests pass (if applicable)
• Manual testing done: non-CUDA targets build without CUDA DT_NEEDED
  entries; CUDA configuration finds and links CUDAToolkit.

Checklist

• I have performed a self-review of my own code
• I have formatted my code using ./scripts/code_format.sh (not applicable;
  no C/C++ source changed)
• I have run pre-commit on the files changed in this PR and all hooks pass
• I have updated the documentation (not applicable)
• I have added tests to prove my changes are effective
• For changes >500 LOC: I have filed an RFC issue (not applicable)

AI Assistance Disclosure

• No AI tools were used
• AI tools were used: Codex helped inspect the published wheel, trace the
  CMake dependency, implement the fix, and run validation. The submitter has
  reviewed the complete diff.