GAP code coverage broken by fork()

[markuspf]

If IO_fork is used while --cover or --profile is active, the resulting profiles are sometimes corrupt. This leads to the io packages' coverage to look even worse than it is.

[fingolfin]

@ChrisJefferson has been working on this

[ChrisJefferson]

I (think) it's now fixed. Wasn't planning on backporti g to 4.10, but we could.

[fingolfin]

Personally, I don't see a need to backport this, but if somebody wants to argue for it, go ahead!

In the meantime, this can then be closed. For the record, I believe PRs #2908 and #2914 are relevant.