Guess a fix to segfault issue

[MichaelChirico]

For #567.

cc @mcol / @struckma -- can you test if this change removes the segfault for you? I'm still hamstrung by being unable to reproduce the segfault outside GHA.

Also cc @krlmlr who filed & followed the orginal Buganizer report & workaround kludge here.

[mcol]

I've run our testsuite with a freshly-compiled covr with your patch applied, and unfortunately it doesn't help: running time was 68 minutes (it's around 7-8 minutes on 4.3.3).

[MichaelChirico]

Alas. IMO the next thing to do is try "bisecting" the package itself to see if it's a certain piece of code that's responsible. Unfortunately for data.table, the issue is most likely related to `[.data.table` which is ~2000 lines long and probably used in >80% of tests... not very friendly for bisection :\