[tests][nice-to-have] Resource Helpers would be more useful as builders

[smklein]

Context: https://github.com/oxidecomputer/omicron/pull/2113/files#r1063578360

The resource helpers might be more useful if they used a "builder pattern", to allow optionally changing features, such as the authentication information across a wide range of NexusRequest objects

[davepacheco]

Yeah, they're a little long in the tooth. At this point, if we were going to do something different, should we be using a progenitor-generated client instead?

[david-crespo]

Yes!

[davepacheco]

Another thing I've wondered about is whether we'd be better off using something like Hurl. I think quite a lot of our happy-path testing and basic error testing could be described in terms of just a bunch of requests and expected responses (and a lot of it could even be automatically generated from the OpenAPI spec). That said, the mechanism we use today also verifies a bunch of other useful properties (like that all 400- and 500-level response codes have a message and error code, that everything has a request id, etc.)