[Epic] Compute Separation — Enable external workers on separate compute

[brettsam]

Goal

Enable language workers to run on separate compute from the host (different VMs or containers), connecting inbound over gRPC. The host becomes a passive acceptor of gRPC streams rather than a process spawner. Workers and their customer payloads become the unit of scale.

The existing host-managed worker model is completely untouched. All new code is parallel and additive.

This is a precursor to a broader worker refactor — once the separated compute model is validated through internal preview, subsequent milestones will build on this foundation to fully decouple worker lifecycle management from the host.

Sub-issues

• [Compute Separation] M1: Extract WorkerChannelBase from GrpcWorkerChannel #11641
• [Compute Separation] PROTO: Add host_configuration_json to StartStream #11642
• [Compute Separation] M2: ConnectedWorkerChannel + channel manager + invocation dispatcher #11643
• [Compute Separation] M2b: ConnectedWorkerFunctionMetadataProvider #11644
• [Compute Separation] M3: Inbound connection acceptance in FunctionRpcService #11645
• [Compute Separation] M3b: host.json delivery via gRPC #11646
• [Compute Separation] M4: DI wiring, configuration, and trigger readiness gate #11647
• [Compute Separation] M4b: Placeholder mode support #11648
• [Compute Separation] M5a: Worker disconnection handling and stateless reconnection #11649
• [Compute Separation] M5b: Structured logging and health checks for external workers #11650
• [Compute Separation] M5c: Integration tests #11651