fix(samples): report failed issue counts and exit non-zero in adk_stale_agent

[chelsealong]

Description

In contributing/samples/adk_team/adk_stale_agent/main.py, process_single_issue() caught any exception from the runner and still returned its normal (duration, api_calls) result. main() then incremented processed_count by len(chunk) regardless of whether any of those issues actually succeeded, so a run where every issue raised could still log Successfully processed N issues. and exit 0.

This PR:

• Changes process_single_issue() to return a success flag alongside its metrics, and sets it to False when it catches an exception.
• Changes main() to only increment processed_count for successes, and to collect failed issue numbers separately, logging them.
• Makes main() raise a RuntimeError if any issues failed, after all chunks have finished (so per-issue isolation within a batch is preserved).
• Updates the __main__ block to track an explicit exit code and call sys.exit(exit_code), so the exception raised by main() actually results in a non-zero process exit status (previously it was caught and logged but the process still exited 0).

Fixes #6520

Testing plan

Added tests/unittests/test_adk_stale_agent_main.py, which imports the sample module (with a fake GITHUB_TOKEN) and:

• Monkeypatches InMemoryRunner, get_old_open_issue_numbers, get_api_call_count, and reset_api_call_count with a fake runner whose run_async always raises (mirroring the issue's reproduction), then asserts main() raises RuntimeError when all issues fail.
• Asserts process_single_issue() returns success=False when the runner raises.

Verified the test fails without the fix: reverted main.py to its pre-fix state (git stash/git checkout) and reran the test — it failed with DID NOT RAISE RuntimeError / ValueError: not enough values to unpack (expected 3, got 2), reproducing the exact bug described in the issue (batch logs "Successfully processed 3 issues" and exits 0 despite every issue failing). Restored the fix and reran; both tests pass.

tests/unittests/test_adk_stale_agent_main.py::test_main_raises_when_every_issue_fails PASSED
tests/unittests/test_adk_stale_agent_main.py::test_process_single_issue_reports_failure PASSED
2 passed

Also ran the broader sample test suite (tests/unittests/test_samples.py) to confirm no regressions: 85 passed.

Ran pre-commit run --files contributing/samples/adk_team/adk_stale_agent/main.py tests/unittests/test_adk_stale_agent_main.py (isort, pyink, addlicense, ADK compliance checks) — all passed.

AI-assistance disclosure

This change was prepared with the help of an AI coding assistant (Claude), with the diff and test evidence reviewed before submission.

[svfat]

I would introduce completed_count because the processed_count value is now controlled by a condition, so the progress log could be misleading

[svfat]

This is not an unexpected runtime failure, some individual issues failing is a normal outcome

[chelsealong]

Addressed both comments: introduced a separate completed_count (incremented for every attempted issue, success or failure) and use it in the per-chunk progress log instead of processed_count, which now only tracks successes. Also removed the RuntimeError for per-issue failures since that's a normal outcome, not an unexpected error — main() now returns a success bool, and the __main__ block sets the exit code from that directly.

[svfat]

Could the author confirm whether this PR, including its review-response commits, was created or updated by an automated AI pipeline? If so, please disclose the tool and confirm that a human has reviewed and tested the resulting changes.

[chelsealong]

Yes — this PR, including the review-response commits (a4384a8, dad579e) addressing the completed_count and RuntimeError feedback, was prepared with the help of Claude Code (Anthropic's AI coding assistant), as disclosed in the PR description. I (chelsealong) reviewed the diff and test output before each push.