Fix unexpected error when testing permutation groups for conjugacy

[hulpke]

This is a work-around for the problem encountered in #3446, that is the code accessing subs.reps[o] is not executed, if it is not in the expected form (true or a list).

In the cases I encountered this resolves the error (and leads to a correct result).

I however do not understand the code (and parameters) well enough to assure that this is simply an omission in the original code, and not a reflection of a deeper problem. Thge code thus issues a warning, if this happens.
I have been running my own calculations with this modification for half a year now and did not encounter any problems, I thus want to offer it to the general system.

Resolves #3446

[ChrisJefferson]

This is also my best guess as to how to fix this too (and I'm also unsure)

[coveralls]

Coverage increased (+0.007%) to 84.521% when pulling 9755bf3 on hulpke:fixes into 77a7c42 on gap-system:master.

[fingolfin]

Thank you!
I have absolutely no idea what's going on in that code, but if both of you think it's plausibel, I fully trust that (and should there be any unexpected regression, we'll deal with that as usual shrug).

[fingolfin]

Should this be backported? Normally I'd say "no, to risky, let's first test it for a while in master", but since @hulpke has been using it for half a year anyway, it seems not so bad... Thoughts?

[hulpke]

@fingolfin
As for backporting, it fixes an issue that has never arisen in 20+ years (and I think I am the only one who encountered it so far) and it gives an error, not a wrong result.

I thus think it is safe to not backport it -- we can always issue a fix later.