Quick fix for cases, where scorch fails on no-cluster samples

[Rexhaif]

Quick fix for #9

[LoicGrobol]

Wow, that was quick, thank you ! Interestingly, those edge cases were apparently never actually documented (at least not in the 2014 paper :-) but your changes are consistent with the reference implementation (for future ref, this is in the ShowRPF function).

Just one thing : can you make these 0s 0.0s instead for type consistency?

[Rexhaif]

Sure, did it
Actually, i was referring to that https://github.com/wikilinks/neleval/blob/68f68414f6a82e3d86761b8d3083e537582bfa7f/neleval/evaluate.py#L281 warning message in neleval library, which also has several coreference resolution metrics implemented