Vertex AI adds Compact HTTP Forwarding and Serving Operations

The new projects.locations.publishers.v1.responses.compact method allows developers to forward arbitrary HTTP requests (streaming and non-streaming) to deployed models, requiring an invoke_route_prefix for path authorization. The servingProfiles resource has also been expanded with standard long-running operation (LRO) methods—get, list, cancel, and delete—improving the management of asynchronous tasks. Finally, the OnlineEvaluator trace scope predicate now explicitly defines its duration filter in seconds.