feat: add Metal backend for macOS and Apple Silicon

[emberian]

This commit adds a new Metal backend to rust-gpu-tools, enabling support for:

• macOS Metal API for GPU compute
• Apple Silicon hardware
• Integration with existing CUDA and OpenCL backends
• Parallel use of multiple backends from the same code

Metal support is behind a feature flag ('metal') and follows the same
API patterns as existing backends for consistent usage.

---

This entire backend (and commit message above the line) was made by claude-code, with ~zero guidance, which I think is pretty neat and am sharing this here.

⎿ Total cost: $5.19
Total duration (API): 18m 54.8s
Total duration (wall): 45m 3.8s
Total code changes: 1254 lines added, 190 lines removed

The actual approach it took towards the conditional compilation isn't what I would have used.

This still needs a bunch of work, but I didn't expect it to get this far. I don't know if I will find the time to pick it up.

[emberian]

ember@nextop rust-gpu-tools % cargo +nightly run --example metal_add --features metal
    Finished `dev` profile [unoptimized + debuginfo] target(s) in 3.85s
     Running `target/debug/examples/metal_add`
Found 1 GPU device(s)
Using device: Apple M2 Max
Running kernel...
Verifying results...
All results are correct!
zsh: segmentation fault  cargo +nightly run --example metal_add --features metal

[BigLep]

@emberian : are you going to take this forward?

[emberian]

@BigLep I am demotivated because my usecase (https://github.com/hellas-ai/hints acceleration) turned out to be slower on GPU than CPU

[vmx]

@emberian Thanks for letting us know. I'll close the PR.