Add an AI review workflow focused on integration test correctness

In integration tests we add or modify, we tend to make the same conceptual mistakes later leading to flakiness:

- not using the test's individual namespace
- invalid testify assertions in Eventually blocks which can only succeed the first time
- not waiting for necessary conditions properly.
- ...

We should produce a proper list with instructions, then have an agentic workflow run on any PR modifying integration tests, looking for these specific mistakes. This should be relatively cheap, as it only runs on diffs, and shouldn't hallucinate too much given the narrow scope.

No files or existing workflow entry points are named. Start by locating the integration-test changes and the PR workflow configuration, then define the promised checklist around individual namespaces, Eventually assertions, and waiting conditions. Done means a focused workflow reviews only relevant diffs and reports these mistakes using agreed instructions.

Written by the indexing model from the issue text.