test(dedupe): TestDedupeDynamo_Unreachable depends on five failures landing within one second #733

Problem. TestDedupeDynamo_Unreachable (tests/integration/dedupe_dynamodb_test.go) depends on wall-clock timing and can fail under load. It makes five failing Reserve calls against http://127.0.0.1:1 and then expects the sixth to be short-circuited, because "five failures in a second open the breaker". When the five calls take longer than a second, the breaker never opens and the assertion at line 247 fails:

Error: "dedupe store unavailable: dynamodb put_item: operation error DynamoDB: PutItem, exceeded maximum number of attempts, 1, ... dial tcp 127.0.0.1:1: connect: connection refused" does not contain "short-circuited"
--- FAIL: TestDedupeDynamo_Unreachable (2.31s)

Evidence (measured).

• Seen once in a local make ci on a busy machine. The tree was main plus files unrelated to dedupe.
• The test took 2.31 s, so the five failures spread over more than the one-second window.
• An immediate re-run on the same tree passed.
• Frequency on GitHub runners is unknown; no CI failure of it has been seen yet.

Context. The test runs with t.Parallel() alongside container-backed integration tests, so a refused-connection dial plus SDK retry setup can be slow when the host is loaded (inferred).

Related: #728 (another timing-sensitive integration test).