Add basic benchmark harness

[ToucheSir]

A revamp of @dfdx's work in #304. The main changes here are running benchmarks and rendering the report on Buildkite instead of GHA, because the former gives us runners with >2 threads.

PR Checklist

• Tests are added
• Documentation, if applicable

[ToucheSir]

Unfortunately it's not possible to compare against master because no such benchmarking code exists there (that's why you'll see this step show up as failed on the latest commit). Therefore, I added an throwaway commit comparing HEAD vs HEAD to show an example of what the output would look like: https://buildkite.com/julialang/nnlib-dot-jl/builds/501#annotation-default.

Posting a GH comment would be far more difficult because of secret management and is best left for a future PR. If anyone finds an easier way to access the buildkite link, however, we could do something like what's currently in place for docs previews.

[darsnack]

This looks good to me, we can add more to the suite or solve the commenting issue separately.

Also, the documentation deployment issues should be resolved, and we can merge this to test that.

[darsnack]

Nice docs appear to be working