Recommend that order of unordered collections is maintained where possible

[benjie]

For legibility and consistency reasons, it's helpful if the order of unordered collections is maintained. For example, an enum describes a set (unordered) of possible values, but it would be irritating if every time you introspected the API or built an SDL for it the enum values were in a different (random) order. This PR recommends that you maintain the initial order that such unordered collections were created with. Theoretically, this means that if you generate SDL from a schema (e.g. via introspection) and then build a new cloned schema from this SDL, generating SDL from this cloned schema (e.g. via introspection) should match the original SDL (i.e. printSchema(schema) == printSchema(buildClientSchema(introspectionFromSchema(schema))))

This PR builds on #1091 and is a change extracted from #1063 as discussed at last night's WG meeting. It does need to be stacked since it also adds the ordering details to Appendix A (originally we thought it might be able to be separate).

The previous PR in this stack is:

• Define Data Collections used in the spec #1091

The next PR in this stack is:

• [RFC] Type system ordering of: object interfaces, directive arguments, input object fields, enum values #1063

[benjie]

This was split from RFC2 PR #1063, so I have marked it as RFC 2 itself.