Relative tolerance instead of absolute tolerance

[0xferit]

As discussed in #350 we are planning to implement relative tolerance and use a standard relative tolerance value to test units instead of using a different absolute tolerance value for each unit.

Benefits are:

• Standardized precision
• Shorter source-code

We can implement it similar to MathNet.Numerics.Precision.Equality

[tmilnthorp]

I think this can be closed now unless anyone has any objections?

[angularsen]

Yup!