Improve Model Target dataset status fidelity

[adamtheturtle]

Follow-up to #3191.

The Model Target Web API mock currently has a simple status lifecycle of processing followed by done, with a small response body. Compare standard and advanced dataset status responses against real Vuforia and update the mock to match the observed response shape and states where practical.

Acceptance criteria:

• Verified fake tests cover status responses before and after completion for standard and advanced datasets.
• The mock returns Vuforia-like status fields, timestamps, and state names.
• Not-found and other status-route errors match real Vuforia where practical.

[adamtheturtle]

Additional clarification after #3191: the current mock-only tests include unknown-dataset status behavior. That is externally observable API behavior and should not stay mock-only unless there is a documented reason.

As part of this issue, audit the status-related mock-only tests and convert unknown-dataset/status-error behavior to verified fake tests where practical. Keep mock-only tests only for internal mechanics such as deterministic processing-time simulation, and document any remaining mock-only status cases in docs/source/differences-to-vws.rst or the test rationale.