Fix bias

[muursh]

Added a nonce when scaling up to help avoid bias. For example scaling the value 77 from the original range 0-100 to the new range 0-255 would have always given 196. 78 would have given 198 and 76 193. The value 194 was impossible with an output. With the addition of this nonce the value 194 is now possible (for several input values). We use a pseudorandom float as the nonce which itself has a pseudorandom scale applied. There is still an issue where the original range is much less than 255. For example scaling from an original range of 40 to a range of 255 will still result in slight bias. Further work is required to fix this for any input. On the assumption we use this library to scale down (which is always the safest behaviour) or to scale up from a range of at least 100 this fix is sufficient.

Tests updated to reflect the new nonce meaning the tests are somewhat less deterministic.