Handle qrel binarization

[JMMackenzie]

We should check how TREC handle qrel binarization and follow the same process, paying attention to types and trying to make things as clear as possible.

[mpetri]

but RBP supports float gain values. do we need to do this?

[mpetri]

actually right now it is a bit of a bug that we cast the value to int and if the max value is > 1 we don't normalize the metric to [0,1] (I guess the RBP paper also doesn't do this)

[JMMackenzie]

Yeah, we should check how to handle - I believe we should allow gain mapping to floats like you suggested, but we'd need to be explicit about it somehow. Perhaps just let the qrels reader handle floats?