fix: preserve terminal failures when result publication fails

[hjcud]

What changed

• preserve a terminal JobNotSatisfiable decision and its original reason code when publishing the failure event itself raises
• log the publication exception with the run, attempt, and original reason instead of letting the outer worker convert it to HANDLER_ERROR:* and retry
• validate and normalize BACKTEST_WORKER_CORRELATION_ID as a UUID before building any other production worker dependency
• add focused coverage for successful terminal failure publication, publication failure fallback, fail-fast invalid configuration, and canonical UUID output

Root cause

A malformed worker correlation ID caused terminal failure-event construction to raise. That secondary exception escaped OrchestratorJobHandler, so the outer worker classified it as a retryable handler error and hid the original REQUIRED_INPUT_UNAVAILABLE decision.

Impact

Terminal binding failures now remain permanent even when their result event cannot be published, and their original reason survives in worker attempt/DLQ handling. Invalid correlation IDs fail at process startup instead of remaining latent until the first result event.

This PR changes only backtest-engine. The backend compiled-plan handoff and root user-data UUID generation remain separately owned and are not modified here.

TDD evidence

Before implementation, the three new regression tests failed with:

• OSError: result sink unavailable escaping the handler
• the unrelated missing runtime-policy setting appearing before correlation validation
• no _required_uuid configuration validator

After implementation:

• python -m ruff check src tests — passed
• python -m mypy — passed, 74 source files
• python -m pytest tests/test_wiring.py tests/test_production.py tests/test_worker.py -q — passed, 61 tests
• python -m pytest -p no:cacheprovider -m "not docker" — 1240 passed, 3 skipped, 168 deselected
• git diff --check — passed

No AWS run, release, DLQ operation, worker restart, ASG operation, or database mutation was performed.

Closes #74

[kcrmin]

검토하고 병합합니다. 요청한 두 가지가 정확히 들어갔습니다 — 발행 실패를 삼키고 원래 reason_code 로 PERMANENT_FAILURE 를 보존하며, _required_uuid 가 다른 의존성 구성보다 먼저 실행되어 잘못된 값이 기동 시점에 걸립니다.

루트 쪽과 맞물리는 것도 확인했습니다. user-data 가 공급하는 파생 UUID 가 str(uuid.UUID(value)) 정규화를 통과하고 값이 바뀌지 않습니다.

i-07a6870a8c4c199dc -> b915ae58-9bfd-5c36-8a36-f28a590403d2   정규화 결과 동일
i-000276dbff2ac0e58 -> 2fec26d4-6d05-51b8-8b58-98938a242337
i-0417d4380dfe35adf -> 33bebc4d-d3af-5755-8cff-dade185b4266

이 저장소는 원래 hjcud 소유이고 제가 병합·후속 작업을 하는 것은 제품 권한자(kcrmin)의 지시에 따른 것입니다 — "hjcud한테 왔다갔다 하는것 없이 그냥 이거 관련된 작업들은 여기서 작업해줘". 왕복을 줄여 INT03 을 오늘 끝내기 위한 조치이고, 이후 이 저장소의 일반 소유는 그대로입니다.

같은 릴리스에 backend #250(체크섬 handoff)과 루트 user-data UUID 수정을 함께 올립니다.