[Cbc] Infeasible problem with multithreading enabled and (multiple or cutlength) option

[John Forrest]

Yves,

Error went away if I increased a tolerance slightly, but I wasn't happy 
with that solution.  Digging deeper I found there was a check in 
CglKnapsackCover to see if we had a valid cover - cover cut exists if 
sum - rhs>tolerance.  sum - rhs was 1.0000029071477456e-07.  tolerance 
was epsilon_ == 1.0e-7 so invalid cut was taken.

At this stage, when doing final checking the tolerance should be 
epsilon2_ which is significantly bigger.

Modified in trunk and stable.

John Forrest