fix(tests): Fix flaky AI conversations tests exceeding Snuba retention

[mchen-sentry]

https://github.com/getsentry/sentry/actions/runs/24205379487/job/70659792593

This flake has been showing frequently and being covered by retries.

The test stores spans into Snuba, then immediately queries them. Sometimes the query returns nothing: assert len(response.data) == 1 fails with 0.

span_to_trace_item() hardcodes retention_days=90 for all test spans. But these tests use before_now(days=107) and similar, so the span timestamps are already past their own retention period at insertion time.

Original tests, days=10..80: ec760c4d8f4
Following tests added days 90..109:

• a0f021e947f
• 5cad1d93b65
• 6c679902c92
• 2afaa854852

Most likely the reason this flakes is that ClickHouse doesn't reject expired data on insert, and relies on a background process to clean up rows past their TTL. Assuming that the test is able to pass when the query runs before that process is able to clean these rows.

https://clickhouse.com/blog/using-ttl-to-manage-data-lifecycles-in-clickhouse

Fixed all 13 tests in this file that had days values >= 90 to within the retention window. Also fixed a duplicate days=30 between two tests.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 206953d. Configure here.}