Fix flaky test [chromium] _ e2e/add-node-dialog.spec.ts:355:1 _ progress advances live from the stream and recovers from a failure with Retry

The merge lane saw `[chromium] _ e2e/add-node-dialog.spec.ts:355:1 _ progress advances live from the stream and recovers from a failure with Retry` fail and then pass on a rerun of the same tree. - Test: `[chromium] _ e2e/add-node-dialog.spec.ts:355:1 _ progress advances live from the stream and recovers from a failure with Retry` - Tree: `5f6d5431b2d6329a2d5afcc1aa5d32f22b164fca` - Lane task: 9696 - Flake record: 01a16ece-ed8e-7951-b39f-1178f4d1f056 Make it deterministic; do not delete it unless it tests nothing.

Fixed the flake: the stream stub could fail the order before the view's first read, and with no failed_stage the command was hidden. The stub now sends failed_stage like the server. Forced race: old stub fails, new passes; 15/15 repeats; quick gate green. PR #6822 queued. Memory: none applied.