More 5'/3' bias edge cases

[kachulis]

Similar to #74 (which has been fixed for the case of only 1 gene being used). The same issue also happens when the number of genes used in the bias calculation is less than 5, the calculation will read past the end of the ratios vector, giving undefined behavior.

[agraubert]

Thanks for pointing this out! I've opened a PR to pull in your fix, and I'll try to review and test when I get a chance #82