For @required fields and Relay Resolvers which might error, we include the field's path (relative to its parent fragment/query) in the generated artifact.

This allows us to avoid maintaining a field path during read traversal, which would require a large number of push/pops of an array as we traverse through the reader AST.

However, with the addition of @throwFieldOnError we also report missing data as field errors, but these errors can occur for any field, and including every field's path in the generated artifact is likely too heavy.

This PR attempt to use an efficient approach where we build up the path on the trailing edge of the reader recursion, which allows us to avoid pushing/popping in the common case where there are no errors. The cost in the happy path is just an additional method call and a few null checks as we exit each level of recursion that impacts the path.

If this approach proves effective, we should be able to adopt the same approach for @requried and resolver errors, simplifying the compiler and making our generated artifacts a bit lighter.