Regression stability + new regression tests

[scarlehoff]

• Add a test for vp-setupfit
• Improve the stability of the regression tests
• Include a theory covmat test
• Generate the data in a dedicated worker
• Set the tolerance with respect to said worker

Fixes #2464

[scarlehoff]

Ok, so the good news is that after many label-unlabel I have now the two samples that produce different results and make the test fail.
The failure seems to be coming from workers in different Azure regions. And there is no way to choose among them. Also, no guarantee that within the same region all workers are the same.

There are two options:

1. Run the regression in our own hardware.
2. Use those two samples to set the tolerances

I think the solution I'll go for is:

• Regenerate the regression in our own hardware (if our github plan allows for it...)
• Use the two samples to select a threshold compatible with the drift with respect to our own hardware.

This should be the most robust option.

Sorry for the spam whoever is still subscribed to this issue/PR but making the CI do it for me in a loop was the lowest-effort way to fish for the errors 😅

[Radonirinaunimi]

Regenerate the regression in our own hardware (if our github plan allows for it...)

For this, I think we can host our own runner in the Nikhef cluster (or somewhere else) as we did with pineko for the regression test.

[scarlehoff]

Ah, fantastic. We should be able to use the same one I think. I was worried the custom runner was a paid feature, I didn't realize we were already using it.

[Radonirinaunimi]

Btw, I will be waiting for this before resuming #2478.