Workflow run benchmarks

[xspronken]

PR containing the first part of the github workflow.
This workflow runs the benchmarks for add and matmul and stores the result in an artefact.
The workflow will not work yet since the previous PR hasn´t been merged yet.

[hodgestar]

Left a few small suggestions for improvements, but looks good.

[hodgestar]

Looking at the benchmark JSON:

  "group": "Add-2-dense-numpy",
  "name": "test_add[2-dense-numpy]",
  "fullname": "test_linear_algebra.py::test_add[2-dense-numpy]",
  "params": {
    "size": 2,
    "density": "dense",
    "dtype": "numpy"
  },
  "param": "2-dense-numpy",

These groups, names, etc are going to be tricky to adjust later, so they should be conceptually clean, not tailored to the current use cases. The name, fullname, params, and param all look sensible. I'm less sure about group.

To me group is defined by "all of the benchmarks in the same group perform the same operation, possible with different parameter". If that is the right definition, then the group names should be things like "add" or maybe "simple-add" or "add-two" (where "two" here refers to the number of things being added, not the size).

I'm fine with this being address in a later PR, but let's at least create an issue on this repo for it now, and keep in mind that we will likely have to throw away or migrate all the benchmark JSON files later if we adjust the groups.

Note that there might be multiple tests that populate the same group (e.g. two different implementations might require rather different implementations). E.g. We could imagine test_add_gpu.

[hodgestar]

I left a long comment about the "group" that each result belongs to, but I'm happy for an issue to be opened to discuss that and for this to be merged now.

[xspronken]

Regarding the group I will create an issue to discuss it.