Match Model Target dataset creation validation

[adamtheturtle]

Follow-up to #3191.

The Model Target Web API mock currently validates only the JSON content type, parseability, required top-level fields, and model counts for standard and advanced datasets. Add verified fake tests against real Vuforia for invalid dataset creation payloads, then make the mock match the observed validation behavior where practical.

Acceptance criteria:

• Verified fake tests cover representative invalid standard and advanced dataset creation bodies.
• The mock matches Vuforia status codes, error body shape, and error targets for those cases.
• Any validation that remains intentionally out of scope is documented.

[adamtheturtle]

Additional clarification after #3191: the current tests/mock_vws/test_model_target_web_api.py::TestMockErrors coverage includes mock-only dataset creation validation cases such as wrong content type, invalid JSON, missing fields, invalid models, and invalid model counts. These were added to close mock branch coverage, but they are externally observable API behavior.

The end state for this issue should be no unnecessary mock-only dataset validation tests: convert representative cases to verified fake tests against real Vuforia, replace mock-only assertions with verified fake equivalents where possible, and document any cases that must remain mock-only because real Vuforia behavior is impractical, unstable, or too expensive to exercise.