I don't really know and the implementer is not reachable. I was under the impression that nBags was the number of bags, meaning the granularity of bootstrap sampling to be done. But one would need to step through code to be really sure.
RankLib Algorithm Parameter Documentation
I am unaware of specifics as to the Random Forest parameters beyond comments in the code. You will note many of the Random Forest parameters are shared by LambdaMART and other tree-based algorithms. I agree that documentation of algorithm parameter specifics would be useful information to add to the RankLib Wiki pages. I will add a feature request to add such documentation.
Indri fails to compile on OS X Sierra
How to specify a stemmer to use at query time?
FeatureManager can produce simple feature distribution statics for models that restrict feature use (e.g. tree models), but is not applicable to models that make use of all defined features (as it is not known how feature weights might correspond to feature usefulness. See feature_stats parameter in the FeatureManager.
This seems correct but I will await comments from the implementer before changing anything.
You can put it anywhere you like as long as the path you use can be found. If you used Maven to compile RankLib, there should be a version 3.5 commons-math3 in your local maven repository. For me, that is my home directory under .m2/repository/org/apache/commons/commons-math3/3.5/commons-math3-3.5.jar.