Story: Hotfix: eventing integration test flakes on Windows CI

The eventing integration test (the write_*_publishes_nats_changed_event family, one per entity) fails intermittently on Windows CI: the NATS notification does not arrive at the observer within the test's 5-second poll budget. The failing entity and job rotate run to run (day_count_fraction_type on 2026-08-22, party_id_scheme on 2026-08-25). This story restores a green Windows CI by making the test resilient to transient infrastructure drops: on timeout, the test re-drives the write (which re-fires the DB notify trigger) within a bounded overall budget, and on exhaustion it reports what it did receive so the loss is diagnosable.

• The eventing integration tests pass on Linux, macOS, and Windows CI.
• The test still asserts the end-to-end chain: a write produces a NATS change notification carrying the written entity id.
• A genuinely broken chain still fails the test (with diagnostics) within the overall budget.

• Resilience lives in the test, not the production chain: the loss windows (listener reconnect backoff 1-5s, NATS publish catch-and-log) are environment transients; the chain itself is correct, so no production code changes.
• Fix the generated template, never a single generated file: one org edit re-tangles and re-generates all 85 tests, keeping the fix and its diagnostics uniform.
• The flake is cross-platform (Windows and macOS, rotating entities), so the fix must not be platform-specific.

Closed at sprint 25 close. Every task is done.