test needed to show that transclusion of $ref into allOf changes the evaluation outcome

see https://json-schema.slack.com/archives/CT7FF623C/p1603727881177900?thread_ts=1603504778.173300&cid=CT7FF623C

Basically, we need a test to show that changing this:

{
  "$ref": "some subschema that generates annotations",
  "unevaluatedProperties": { ... },
}

to this:

{
  "allOf": [
    < that same subschema that generates annotations, included directly here... >,
    "unevaluatedProperties": { ... }
  ],
 
}

...does not always yield the same result, sibling keywords in a subschema can see annotations produced by earlier-evaluating siblings, but subchemas under allOf siblings cannot see each other's annotation results.

Therefore, this is not a reliable mechanism of transcluding/bundling referenced subschemas into the same schema.

cc @Relequestual

Start by reviewing the repository's existing tests for $ref, allOf, annotations, and unevaluatedProperties, then consult the linked Slack discussion for the intended cases. Add a regression test demonstrating that the two schema forms can produce different evaluation outcomes.