Stabilize certificate rollback partition test

[Amaury Chamayou (achamayou)]

Summary

Use the test network's election-derived timeout while waiting for commit levels to converge after healing the partition in test_rolled_back_node_certificate.

This is deliberately scoped to this call site. The generic 3-second default remains unchanged for tests that do not intentionally tear down node-to-node connectivity.

Why

The first attempt of CI run 31189649906 failed after the certificate-renewal partition was removed:

• node 0 rolled back its conflicting view-6 suffix and stored the view-7 suffix through 7.96;
• node 0 reported 7.94 committed while nodes 1 and 3 reported 7.96 for the full 3-second wait_for_node_commit_sync() default;
• the network's configured election timeout was 4 seconds;
• just after that 3-second wait failed, node 0 reopened node-to-node connections, started view 8, and became leader with 7.96 in its log.

Deleting the iptables rules and reaching primary unanimity do not guarantee that node-to-node transport has recovered immediately. This is the same class of timing behaviour described in #7687 and handled elsewhere by #7601.

The preceding wait_for_commit() in this test already uses network.election_duration * 4. Applying the same bound to commit synchronization gives transport reconnection and any resulting election time to complete, without adding an unbounded wait or weakening assertions.

Attempt 2 passed unchanged, including the partitions step, confirming that the original failure was intermittent rather than a deterministic consensus regression.

Validation

• scripts/ci-checks.sh
• cmake -S . -B build -GNinja -DCMAKE_BUILD_TYPE=Debug
• cmake --build build
• ./tests.sh --timeout 360 --output-on-failure -L partitions -C partitions (passed in a user network namespace with NET_ADMIN)

[Copilot]

Pull request overview

This PR stabilizes the test_rolled_back_node_certificate partitions test by aligning the post-partition commit-synchronization wait with the test network’s election-derived timing, reducing intermittent CI failures when node-to-node transport recovery lags behind primary unanimity.

Changes:

• Extend wait_for_node_commit_sync() timeout at the rollback verification point to network.election_duration * 4 (instead of the 3s default).

Custom instructions used:

• .github/copilot-instructions.md
• .github/instructions/reviewing.instructions.md