Bug fixes

• Vertex AI endpoints, models, deployment resource pools and notebook runtime templates now stay up to date with GCP on every CMDB run. Previously some kept the details captured when they were first discovered, and endpoints kept details removed in GCP, such as deployed models after the last undeploy, or request-response logging after it was turned off.
• Guardrails now picks up these Vertex AI changes in real time: new models (including Model Garden deploys) and deployment resource pools, models being deployed to or undeployed from an endpoint, and updates or deletions of endpoints, models, model versions, deployment resource pools and notebook runtime templates. Deploying from Model Garden, or creating a notebook runtime template with gcloud or the API, no longer adds an unnamed resource to Guardrails.