refactor: simplify GHCR deploy poller

[kmatsunami]

Summary

• reduce the GHCR deploy poller to the minimum env-driven flow: resolve digest, compare, deploy, smoke check, rollback
• remove poller state tracking, dry-run support, and the macOS launchd sample to keep the deploy contract small
• rewrite deploy poller tests and align README/systemd/operations docs with the simplified behavior

Test Plan

• /Users/kentoku.matsunami/Documents/GitHub/comic_crawler/.venv/bin/python -m unittest tests.test_ghcr_registry tests.test_deploy_poller tests.test_check tests.test_status tests.test_runner -v
• /Users/kentoku.matsunami/Documents/GitHub/comic_crawler/.venv/bin/python -m unittest tests.test_source_drift tests.test_sources tests.test_update_classification tests.test_check tests.test_status tests.test_watchlist tests.test_runner tests.test_migrate_v2 tests.test_backlog

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ce4e9a68a3

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Skip rollback when prior env digest is not deployable

run_once now always rolls back to the pre-run COMIC_CRAWLER_IMAGE_REF after any deploy failure, but this ref can be an initial placeholder (the docs allow sha256:replace-me before first deploy). In that case, a failed first deploy immediately triggers a rollback to an invalid digest, which reliably fails and obscures the real failure path instead of falling back to manual intervention. Rollback should be conditional on having a known-good previous digest/reference.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Preserve original deploy error when rollback fails

The rollback call in run_once is no longer wrapped in its own try/except, so if rollback fails (for example, old image pull/up fails), the function raises only the rollback exception and drops the original deploy/smoke-check error context. This regresses troubleshooting and alert quality compared to the prior behavior that reported both failure stages. Catch rollback exceptions and re-raise with both causes (and a rollback-failed notification) to keep incident diagnosis actionable.

Useful? React with 👍 / 👎.