The AI Tool spec that checks that the tool runner will not timeout on slow HTTP request is flaky. In this PR we attempt to resolve the flakiness by: Ensuring stub_request runs before the request Increasing the timeout for CI env