fix: update format for servers.url and links.operationRef in v3.1 schema

[MikeRalphson]

uri-reference is not the right format for these properties, as they can contain unescaped { and } characters. As @jdesrosiers points out, they are not uri-template either, as they do not support RFC6570 URI templates in full.

[handrews]

can't operationRef have embedded runtime expressions in addition to the regular templating?

[MikeRalphson]

@handrews yes, but does that mean it requires a new format value?

[handrews]

@MikeRalphson it depends on what you want tooling to be able to do with the format. If all you're going for is "the old format was wrong and we should have a not-wrong format", then pretty much anything you want is fine. If you want schema tools to be able to recognize exactly what sort of templating is happening and validate that (either within JSON Schema if the format-assertion vocabulary is in use, or as a post-schema-validation step if the format-annotation vocabulary is in use, which it is the default), then your format value(s) need to be specific enough to convey that information.

It's hard for me to judge whether this format value is what we'd want in either or both situations without knowing the expectations for implementation. This, btw, is part of the problem of format, as we saw with decimal — it seems to encourage underspecification and over-expectation. In other words, format consistently over-promises to schema authors while allowing/encouraging implementation to under-deliver.

But really, if all you want this to do is indicate "uri-reference with balanced { and } allowed" then sure this is fine. I did not notice yesterday (COVID-brain) but now I see that in PR #3234 you are referencing the Paths Object for "usage" (rather than syntax?) and not the Operation Reference. Does the operation reference allow syntax that is not allowed in the Paths Object? I'm definitely a little unclear on how runtime expressions fit into the templating for the operation reference - I need to study that when I'm feeling better.

[handrews]

See also #3256 (Formally define templating behavior)

[handrews]

While I agree that we need to do something like this, I'm going to close this particular PR so that:

• We can merge Drop uri-reference format for template URLs #3455 to at least get the wrong format out of the schema
• We can resolve the question of what to call the format in Add templated-url format to the registry #3234 and/or Define formats for URL templating with and without runtime expressions #3530
• We can resolve the question of how many formats we need in Define formats for URL templating with and without runtime expressions #3530

We can open a new PR for this once the number and name of formats is resolved.