Fix replica drift convergence test race

[remdui]

Cause

The merged main CI exposed a timing race in ReplicaControllerTest.followerConvergesThenBacksUpAndRepairsFilesystemDrift.

The test waited only until config.yml had been materialized as generation 2, then immediately asserted that the host had reconciled. Production applyRemote() intentionally performs these steps in order:

1. materialize authoritative files;
2. reconcile the host graph;
3. update the applied generation/LKG/status.

A runner can therefore observe the file between steps 1 and 2, making the test flaky even though the controller behavior is correct.

Fix

Wait for the completed convergence contract instead of the intermediate filesystem write:

• config.yml == "two"
• host reconciliation has occurred
• ReplicaStatus.appliedGeneration == 2

Then assert the controller is READY before introducing manual drift.

No production code changes.