Task: Implement Hotfix: eventing integration test flakes on Windows CI

Make the eventing integration test resilient to transient infrastructure drops on Windows CI: re-drive the write on timeout (within a bounded overall budget) so a notification lost to a listener or NATS reconnect is recovered, and report received traffic on exhaustion. Change only the test; do not change the production chain.

• The test passes locally (Linux) with the retry path exercised.
• The test still fails when the chain is genuinely broken.
• No production code changes.

Diagnosis (2026-08-25): the failing test is write_party_id_scheme_publishes_nats_changed_event; the read_latest_party_identifier_by_id line in the CI output was a passing test that ran immediately before. The CI log shows the failure ~4.4 s after the test started — the full 50x100 ms poll budget burned. The same binary ran ~25 sibling eventing tests (before and after) green, so the loss is a per-test transient, not a systemic outage. The same template failed on day_count_fraction_type on 2026-08-22 (clang-release) and party_id_scheme on 2026-08-25 (clang-debug): the failing entity and job rotate — a timing/health flake, not an entity bug. The test was added 2026-08-09; Windows CI has been red or cancelled since, so "worked before on Windows" predates the test.

Verified loss windows in the chain (no fix here, they are environment transients): the postgres listener's reconnect path (consume_input failure -> 1-5s doubling backoff -> re-LISTEN) loses notifications in flight during the outage; NATS publish failures are caught and logged silently by publish_entity_event and the event_bus. Both are invisible to the test — it only sees an empty poll.

Fix: per-test re-drive. Write -> poll ~2.5 s; if no matching message, write again (a new bitemporal version row with the same code re-fires the AFTER INSERT trigger; the observer matches on entity_ids). Bound at 4 attempts / ~10 s total. On exhaustion, log what the observer did receive (decoded entity-change events, any NATS messages) so a future failure names the loss.

Result (2026-08-25): implemented and verified. The org template (ores.cpp.eventing-integration-test.nats_integration_test.org) now polls in bounded re-drive attempts: write, poll ~2.5s, re-drive on empty (a new bitemporal version row re-fires the AFTER INSERT notify trigger), bounded at 4 attempts; on exhaustion it logs the observer's final snapshot so a genuinely broken chain is diagnosable. Regenerated all 85 eventing test files across iam/refdata/reporting/trading from the template. Verified on Linux: 62/62 refdata eventing tests green (186 assertions). First run after a cold build took 14.5s and recovered via re-drive – the exact transient the fix targets; re-runs take 0.3-1.7s. macOS failures (instrument_code, counterparty_contact_information, day_count_fraction_type) confirmed the flake is cross-platform; pattern captured as project memory eventing_integration_tests_traffic_light.

Manual QA scenarios (scaffolded via compass add test_scenario, run through the QA Validation Runner panel) that verify this task. Link new ones here as they're created; the scenario doc itself links back via its "Verifies task" field.

Implemented and verified 2026-08-25. The org template ores.cpp.eventing-integration-test.nats_integration_test.org polls in bounded re-drive attempts: write, poll ~2.5s, re-drive on empty (a new bitemporal version row re-fires the AFTER INSERT notify trigger), bounded at 4 attempts. On exhaustion the test logs the observer's final snapshot (decoded entity, subject, size) so a genuinely broken chain is diagnosable. Regenerated all 85 eventing test files across iam/refdata/reporting/trading; no production code changed. Verified locally: 62/62 refdata eventing tests green (186 assertions); a cold chain right after a build recovered via re-drive (14.5s first run) where the old single-budget poll would have failed. The flake is cross-platform (Windows and macOS), so the template fix is the common strategy; the pattern is captured as memory eventing_integration_tests_traffic_light.