Improve testing on ParallelTestRunner

[luraess]

Following up on switching to ParallelTestRunner.jl as testing infra in #876 , there are several points that were not fully resolved such as:

• conflicts in argument parsing between the custom mechanism implemented for ReTestItems and ParallelTestRunners, making it impossible to e.g. select the number of workers else than from setting the CPU threads;
• command line options https://juliatesting.github.io/ParallelTestRunner.jl/dev/#Command-Line-Options would have no effect;
• GPUArrays test were hardcoded while we actually want to dynamically discover them to avoid missing new additions/deletions/renaming

We now only miss the actual way of selecting tests based on some pre-defined categories as it "used to be":

"core", "device", "hip", "external", "gpuarrays", "kernelabstractions", "wmma", "enzyme"

For this reason, I partly renamed the tests as one could now pass either or combination of those terms in order to trigger similar behaviour as it used to be prior to #876 , e.g.,

import Pkg
Pkg.test("AMDGPU"; test_args=`--jobs=4 core device`)

without clashing with ParallelTestRunner setup. If there is a better way to address this last bit without too much of a machinery, then we could give it a shot.

In any case, feedback welcome!

[giordano]

Very minor change:

Suggested change

	jobs_pos = findfirst(arg -> startswith(arg, "--jobs"), ARGS)
	jobs_pos = findfirst(startswith("--jobs"), ARGS)

Also, you aren't really interested in the position, so you can also do something like

if !any(startswith("--jobs"), ARGS)
   push!(ARGS, "--jobs=4")
end

(similarly below)

[giordano]

Just for reference, there's interest in having a way to specify serial tests: JuliaTesting/ParallelTestRunner.jl#77. I discussed with Valentin a way to implement it, but haven't found the time to do it (my need for this feature shifted, at least in the short term, but I think it'd still be nice to have, and I may still need it later)

[luraess]

Thanks for the heads-up. I will leave things for now, and happy to move to the new implementation when available for use. I'll leave a comment in the file with link to the issue.

[gbaraldi]

Thanks for the improvements. LGTM

[luraess]

Thanks. I will wait until CI runs and then merge.