[Compute Separation] M5a: Worker disconnection handling and stateless reconnection

[brettsam]

When a connected worker's gRPC stream ends, the host must clean up gracefully. Reconnecting workers get a fresh channel (stateless — new workerId).

Dependencies: M4 (#11647)

Acceptance Criteria

• FunctionRpcService publishes WorkerErrorEvent when external worker stream ends
• WorkerConnectionService removes channel from IConnectedWorkerChannelManager on disconnect
• In-flight invocations are drained before channel removal
• HostJsonContentProvider.Reset(clearCache: true) called on worker disconnect
• Reconnecting worker with new workerId gets a fresh ConnectedWorkerChannel
• Unit tests for disconnect + reconnect scenarios