Refactor benchmarks

[christiangnrd]

Benchmarks are now a benchmark test suite that tunes and warms up.

To run all benchmarks on CPU (from AcceleratedKernels/benchmark):

$ julia --project runbenchmarks.jl

To run all compatible benchmarks on Metal (assuming you have added Metal.jl to your benchmark environment):

$ julia --project runbenchmarks.jl --Metal

To run only the accumulation benchmarks on CUDA:

$ julia --project runbenchmarks.jl --CUDA accumulate

[anicusan]

This is great, thanks @christiangnrd ! We should add a performance regression tester on pull requests based on this