Allow different $schema values in same document if different resources

[handrews]

Currently, you can put $schema in a resource root object, which means you can have multiple $schema keywords in the same document. Currently we say that schema authors SHOULD only have $schema in the document root, and we say that if multiple resources in the same schema have different $schema values, the meta-schema is undefined. There's a CREF giving some of the backstory.

Before the schema resource concept was introduced, and tied to the updated $id keyword, there was concern about allowing $schema to be dropped into any random schema object.

But now we have two types of schema objects: schema resource root objects, and schema objects within a resource. The document root object is a special case of the schema root object.

Schema resource roots are easily detected ($id means its a schema resource root), and we already do special processing of them. Allowing the $schema to change there makes sense, as we want schema resources to behave the same whether they are linked with $ref (in which case you can definitely change $schema) or not (which is currently undefined).

I think we should go ahead and define this behavior. It also fits in with #849 (clarifying and consolidating the schema loading process), where we will talk about how to load embedded schema resources within a single document.

[Relequestual]

Splitting up $id into two use cases ( now with $anchor) was a great idea.
Yup. I'm all for the above!

[karenetheridge]

It should also be possible (thank you $recursiveRef!) to make this restriction explicit in the metaschema with a rule "if the $id property appears, $schema is allowed; else, $schema is prohibited".

[handrews]

@karenetheridge we've generally avoided enforcing those things in the meta-schemas because they are technically not forbidden. They just have "undefined" behavior.

Due to $recursiveRef and $recursiveAnchor, though, you can write your own strict meta-schema with the same standard vocabularies to enforce whatever makes sense to you.

I agree that this is a gray area, (as opposed to people wanting us to, say, forbid properties unless type is object), and I could be convinced that we should enforce resource roots vs subschemas in the meta-schema. But I'm not yet convinced of that. I kind of want to see what people do in practice with making custom meta-schemas now that we've made that easier and more effective.