I've reviewed sections 2 and 5 and applied various editorial changes to fix issues, make it consistent with the behavior of the implementation in graphql@alpha, make it more consistent with the terminology of the spec, add some additional examples, use definitions for terms, etc.

One key change is that we don't evaluate the variable's value to determine two variables are equivalent, instead we simply require that they come from the same definition.

I've also re-worked some algorithms that were trying to cover arguments and variables with the same code to instead split the handling of that code. Where algorithms do handle both, I've renamed some variables to make this clearer.

I've also tried to consistently use "fragment variables" for the variable definitions, and "fragment arguments" for the arguments passed to the spread; previously these were used a little more interchangeably. I've also used "named fragment spread" rather than just "fragment spread" in many places to make it clearer this only applies to named fragments (not inline spreads).