docs: clarify one-based step attempts

[zhongkechen]

Summary

• document StepContext.getAttempt() and TestOperation.getAttempt() as one-based
• align context implementation Javadocs with the existing runtime and plugin contracts
• add an integration test proving a retried step observes attempts 1 and 2

Testing

• mvn spotless:apply
• mvn -pl sdk-integration-tests -am -Dtest=RetryIntegrationTest -Dsurefire.failIfNoSpecifiedTests=false test

Related shared-docs PR: aws/aws-durable-execution-docs#267

[github-actions]

Claude AI review

This is a documentation-only PR (plus one new integration test) that corrects getAttempt() Javadocs from "0-based" to "1-based" across StepContext, StepContextImpl, DurableContextImpl, and TestOperation.

Verification performed:

• Confirmed against StepOperation that FIRST_ATTEMPT = 1 and retries use attempt + 1, so attempts genuinely start at 1 — the doc correction is accurate, not a new inaccuracy.
• Confirmed RetryStrategy.makeRetryDecision and RetryStrategies (fixedDelay, linearBackoff, exponentialBackoff) already document/treat attempt as 1-based, so this aligns previously inconsistent Javadoc with the rest of the codebase.
• TestOperation.getAttempt() previously said "defaulting to 0" while the code defaulted to 1 — the new comment ("defaulting to 1") now matches actual behavior.
• The new RetryIntegrationTest.testStepContextAttemptNumbers_ShouldBeOneBased correctly exercises RetryStrategies.fixedDelay(2, Duration.ofSeconds(1)): attempt 1 fails and triggers a retry (1 >= 2 is false), attempt 2 succeeds (matches FIRST_ATTEMPT/retry-increment logic), and assertions (List.of(1, 2)) match the expected attempt sequence. Imports are correctly ordered and used; no Thread.sleep() is introduced (uses the test runner's simulated advanceTime()/runUntilComplete()).
• No production behavior, public API signatures, or serialization changed — all non-test changes are Javadoc comment edits only.

No correctness, API-compatibility, checkpoint/replay, concurrency, or lifecycle issues were found in this diff. Residual risk is minimal: this PR does not update other doc pages (e.g., docs/core/steps.md) that don't currently mention attempt numbering, but that is out of scope for this diff and does not introduce any inconsistency.

Reviewed commit fabab980a04b48ef00156f6a8946fe8ebd417950. Workflow run

[github-actions]

Codex AI review

No actionable findings. The changes align the Javadocs with existing one-based attempt semantics, and the integration test covers retry attempts 1 and 2. Residual risk: tests were not executed under the review constraints.

Reviewed commit 67d4b934e7467b152c88ae09798947f28e4f0e01. Workflow run