Add better logging

[odow]

Each of the algorithms could use some debug logging. This would help highlight numerical issues, or cases in which we iterate forever.

Xavier says:

If we want the community to embrace MOA, I feel that we need to be cautious and perhaps refine the implementations so that they are comparable to the authors' native implementations. For example, I recently tried to solve an instance with TambyVanderpooten, MOKP$_$p-6$_$n-30$_$1.dat \(https://github.com/tambysatya/TamVan19/tree/master/Instances/MOKP), which is on Tamby's repo and presented as solved by their algorithm in the paper. It does not run on the MOA implementation (nothing after 24 hours)

24 hours is ridiculous. It's probably a numerical issue and we're just iterating forever on the same box.

[xgandibleux]

Given that these instances are around (or larger than -to make the challenge funny-) the largest instance sizes reported in the authors’ paper (i.e. 6 objectives and 30 variables), long computation times should be expected (ps: 24 hours is « generous » but not that long :) ; the extreme case I recently encountered was 7 full days on 1 instance, and MOA ended with a consistent result…during these 7 days, I often thought to be in presence of a numerical issue, and to kill the process.