fix #386

[gares]

fix #386

[proux01]

@gares could you try proux01@bc02eab

[gares]

done, but I think there was a "race"

[proux01]

Indeed, my bad, I rebased the branch, relaunching should now work.

[gares]

Now it works, but the OOthm fails (as in Coq's CI for your PR).
So it is good enough to bench mathcomp

[proux01]

Bench is a bit disappointing, the last commit use primproj for class->mixin builders actually makes things a bit slower:

• mathcomp
  • before: 13:30, 1772552 kb
  • after: 13:55, 1756420 kb (+3%)
• analysis
  • before: 12:46, 1771820 kb
  • after: 13:13, 1804272 kb (+4%)

[CohenCyril]

I suspect a proper use of universe polymorphism will have a better impact.