Validate @defer directive in schema matches GraphQL spec definition

[calvincestari]

This is in reference to an interesting bug reported on Discord - https://discord.com/channels/1022972389463687228/1174776645358465114/1174776645358465114.

tl;dr - AppSync appears to support a version of @defer and it's in schemas downloaded from AppSync. Our JS frontend manually adds the @defer directive to schemas so it can be used in operations. The directive from AppSync differs in that it allows it to be used on FIELD locations only.

The work here is to make sure that we're not adding a duplicate directive which would cause schema validation to fail. We also need to make sure that if one exists it's on the same locations and supports the same arguments otherwise we should fail our validation.

[harryblam]

For visibility, I'm currently getting the below error when generating my graph in version 1.7.1, (whereas in 1.7.0 it was working fine). It sounds like this PR may fix that.

Error: JavaScriptError: GraphQLSchemaValidationError-There can be only one directive named "@defer".-GraphQLSchemaValidationError@

[calvincestari]

For visibility, I'm currently getting the below error when generating my graph in version 1.7.1, (whereas in 1.7.0 it was working fine).

@harryblam that's correct because some of the defer work has been merged into main.

It sounds like this PR may fix that.

There isn't a PR to change this yet, this is just an issue to track the problem.

Are you also using AppSync though? If you are this error will remain even after it stops adding a duplicate directive. The reason for that is the check I'd implement would be to ensure that any pre-existing @defer directive must be on the same locations (FRAGMENT_SPREAD and INLINE_FRAGMENT) and support the same arguments (if and label). If the pre-existing directive differs from those codegen will abort because our codegen engine is not built to support defer in any other way.

Once the directive is merged into the GraphQL spec, AppSync is going to need to change in order to be spec compliant.

[calvincestari]

I recently made a change to not duplicate the directive, which may be correct for the long-term too, but I still feel we need a check to ensure that any existing defer directive must comply with the RFCs specification of the directive.