Initial support for Metal GPUs

[ronisbr]

Hi!

This pull request has an initial support for Metal GPUs. It was tested in a M1 MacBook. The tests are passing:

❯ julia runtests.jl
      From worker 3:    A = Float32[2.3, 2.3, 2.3, 2.3, 2.3, 2.3, 2.3, 2.3]
Test Summary: | Pass  Total  Time
CPU           |    1      1  8.8s
┌ Warning: No CUDA devices available, skipping tests
└ @ Main ~/tmp/DaggerGPU.jl/test/runtests.jl:46
Test Summary: |Time
CUDA          | None  0.0s
┌ Warning: No ROCm devices available, skipping tests
└ @ Main ~/tmp/DaggerGPU.jl/test/runtests.jl:70
Test Summary: |Time
ROCm          | None  0.0s
Test Summary: | Pass  Broken  Total  Time
Metal         |    1       1      2  4.2s

[jpsamaroo]

Thanks!