Fix sync pruning safety and full-width page layouts

[jwmossmoz]

Summary

Sync safety

• Require every configured Taskcluster pool fetch to succeed before absence checks or sync success
• Fail Puppet syncs on unreadable inventory files and persist the failed sync log after rollback
• Include Windows inventory freshness and presence in reconciliation pruning decisions
• Add SQLite-backed regression tests for all three safety paths

Full-width page layouts

• Remove the remaining page-root width caps from Fleet, Alerts, Insights, Pool Health, and Worker Detail
• Keep Overview full width and let the legacy Map/Workers redirects inherit the Fleet layout
• Preserve intentional width constraints on controls, cells, and dialogs

Reproducible Docker builds

• Add .dockerignore exclusions so local node_modules, dist, Python bytecode, and Terraform provider caches cannot contaminate production image layers

Testing

• docker build -t hangar-pr77:local -f backend/Dockerfile .
• Ran the production image against isolated Postgres and verified /api/health plus a seeded Worker Detail API
• Ran Dockerized Playwright/Chromium 1.61.1 at 2560×1188 across all 12 route variants: every route returned HTTP 200, main/content widths were 2336 px, and every right gap was 0 px
• Ran python -m unittest discover -s tests -v inside the production image: 3 passed
• Source assertion confirms no top-level page root retains a max-w-* class
• autoreview: clean (no findings, 0.91 confidence)

Known baseline issue

• Targeted ESLint reports existing errors in unchanged Alerts, FailureInsights, and Pools logic; the layout commit changes only five class strings.