test(bedrock): stabilize the cache-point-after-document integ test

[opieter-aws]

Description

The test_a_cache_point_after_a_document_is_accepted[pdf] integration test intermittently failed with Timeout (>90.0s). The global pytest ceiling is 90s, but the SDK's default throttle-retry backoff can reach ~124s on its own (4→8→16→32→64), and the [pdf] case compounds it by sending a ~100 KB document and asking the model to Summarize. — a slow generation. A throttled or slow run therefore outlasts the 90s ceiling before the SDK's own retry can finish.

Two changes address both causes: a per-test @pytest.mark.timeout(300) gives the throttle backoff room (matching other retry-heavy Bedrock tests in the suite), and the trailing instruction becomes Reply OK. so the common path returns in seconds. The test only asserts the request is accepted (stop_reason == "end_turn") — it verifies the auto cache strategy places a cache point around a document in a shape Bedrock accepts — so the terse prompt keeps coverage identical while cutting generation time.

Related Issues

None — surfaced by a flaky CI integration-test run.

Documentation PR

None — test-only change.

Type of Change

Bug fix

Testing

Ran the target test against real Bedrock in us-east-1, both csv and pdf parameters, four times: consistently passes in 5–11s (previously timing out at >90s), reaching stop_reason == "end_turn". ruff and mypy clean.

• I ran hatch run prepare

Checklist

• I have read the CONTRIBUTING document
• I have reviewed and understand every line of code in this PR, including any generated by AI tools, and I can explain why it works
• My change is focused and reasonably small; I have split unrelated work into separate PRs
• I have added any necessary tests that prove my fix is effective or my feature works
• I have updated the documentation accordingly
• I have added an appropriate example to the documentation to outline the feature, or no new docs are needed
• My changes generate no new warnings
• Any dependent changes have been merged and published

---

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[github-actions]

Assessment: Approve

Well-diagnosed, minimal test-only fix. The two changes cleanly address both root causes of the flake — the raised per-test timeout gives the SDK's throttle backoff room, and the terser prompt removes the slow-generation variable — without altering what the test verifies.

Review notes

• Timeout marker: @pytest.mark.timeout(300) is consistent with the other retry-heavy integ tests in the suite (memory/* at 300, test_mantle_routing at 600), and the inline comment documents the rationale.
• Prompt change: Coverage is preserved — the test only asserts stop_reason == "end_turn" (that the auto cache strategy produces a Bedrock-accepted shape), so Reply OK. keeps the assertion meaningful while cutting generation time. It also matches the existing convention already used in the sibling test_a_caller_ttl_does_not_conflict_with_the_tools_ttl test.
• Scope: No public API impact, no duplicate functionality, well within size guidance.

Nice, focused fix with a clear explanation of both root causes in the description.