Vertex AI: Task Stores, Speech Transcription, and Model Proxies

The Reasoning Engine's A2A (Agent-to-Agent) task framework has been significantly updated; tasks can now be managed via a new taskStores resource path, decoupling task persistence from the execution engine. For Generative AI workflows, a new transcriptionConfig in GenaiVertexV1beta1GenerationConfig enables integrated speech recognition (ASR). Additionally, the API introduces publishers.v1.responses.compact, a method for forwarding arbitrary HTTP requests to deployed models, supporting both streaming and non-streaming use cases. Agent monitoring has also been broadened, with AgentResource now supporting runtimes on Cloud Run, GKE, and GCE. Finally, a new QUEUED status has been added to interactions to represent tasks waiting for off-peak capacity.