Seed RNG to a fixed value at start of the test suite

[fingolfin]

Resolves #1437 (maybe?)

CC @benlorenz

[codecov]

Codecov Report

Merging #1454 (bd92d69) into master (d82760d) will decrease coverage by 0.14%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #1454      +/-   ##
==========================================
- Coverage   84.73%   84.59%   -0.14%     
==========================================
  Files         110      110              
  Lines       29361    29361              
==========================================
- Hits        24879    24838      -41     
- Misses       4482     4523      +41     

see 8 files with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[benlorenz]

I am in favor of this, the number of tests for various testsets do look like that the rng has the desired effect. But there are still some huge differences in runtime between 1.10 and nightly:
1.9

Generic.UnivPoly.adhoc_binary_operations | 5313   5313  0.3s

1.10

Generic.UnivPoly.adhoc_binary_operations | 5313   5313  0.5s

nightly

Generic.UnivPoly.adhoc_binary_operations | 5313   5313  1m32.8s

For older CI runs of this testset the number of tests varies from run to run but here it is constant across julia versions except for 1.6. Maybe running this with the output from #1432 would be good.

[fingolfin]

We merged #1432. Shall we merge this one, too, @thofma ?

[thofma]

I am not too big of fan of fixed seeds, but I can see that this might be a good idea in view of our recent CI problems.

[fingolfin]

Yeah we can always un-fix it in a future version (but then we should at least print the initial seed for reproducibility)