So that they can be shared with MP OpenAPI 4.0

Also add a validation result matcher so we get better test failure
messages.

Once the app has been deployed, we expect the messages to have been
logged, so we don't need to wait for them. This has no effect if the
test passes, but makes it run much more quickly if the expected messages
are not emitted.

OpenAPI 3.1 is fairly similar to 3.0, so we re-use a lot of the same
validation logic.

As well as adding the new rules, we needed to:
- refactor reference validation, as references need to be validated
  differently
- Change how we select which validator to use so that its based on the
  OpenAPI version specified in the document rather than on which version
  of the feature is running (ensuring mpOpenAPI-4.0 can still validate
  OpenAPI v3.0 documents if the user has requested that)
- Make the 3.0 Schema validation code work with the mpOpenAPI-4.0 data
  model which has breaking API changes. The 3.0 validation needs to work
  with all versions of the model.

We were testing for each of the problem messages we expect, but we
weren't testing for the top message which says "The following errors
were found:"

Copy the existing OpenAPI v3.0 validation FAT tests and adjust them for
the new validation rules in v3.1 and run them against OpenAPI v3.1
documents.