Make Metal intrinsics device-only

[napes]

Summary

This marks the Metal barrier and version intrinsics as device functions so host-side AOT/codegen does not emit AIR/Metal device-only references for these APIs.

The change keeps the existing docstrings attached to the public functions while moving the actual intrinsic bodies behind @device_function.

Why

Including Metal in a host sysimage can expose device-only AIR/Metal symbols to the host linker. These intrinsics are only valid for Metal device compilation and should not create host link obligations.

Validation

• Rebased onto latest JuliaGPU/Metal.jl:main at 224d0e29.
• git merge-tree --write-tree upstream/main HEAD completed successfully, producing merge tree cfb1bf62cfd5171e8eeeac7f0e80850bf00fa866.
• julia --project scripts/check_host_device_intrinsic_regression.jl passed from the repro environment with required failures: 0 and candidate leaks: 0.

Notes

The repro environment resolved Metal v1.10.0 from /Users/ian/Documents/Metal.jl and GPUCompiler v1.22.3 from JuliaGPU/GPUCompiler.jl#main.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 83.63%. Comparing base (224d0e2) to head (769c004).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #849      +/-   ##
==========================================
- Coverage   84.00%   83.63%   -0.37%     
==========================================
  Files          69       69              
  Lines        4400     4394       -6     
==========================================
- Hits         3696     3675      -21     
- Misses        704      719      +15     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.