ci(drift): cross-repo guards — docs selectors ↔ chart labels; preflight thresholds ↔ docs sizing

[saadqbal]

Follow-up to #218 (merged 2026-06-08).

#218 added intra-repo drift checks (backend-host parity across the three installer files + the workload-name contract between summary.sh/diagnose.sh and the chart render). Two cross-repo surfaces are still unguarded:

1. Selector drift — the Mintlify docs repo documents kubectl -l app=… selectors that must match the chart's pod labels. This is the original app=manager bug surface: a label rename in the chart silently breaks the documented commands.
2. Sizing drift — the preflight RAM/CPU/disk thresholds (scripts/lib/preflight.sh, raised in fix(installer): hard-gate low RAM and measure the container-runtime's view #217) must stay in sync with the sizing numbers in docs (environment-setup / system-requirements).

check-drift.sh is single-repo, so this needs a cross-repo guard — likely a scheduled workflow that checks out both repos, or a shared check in tracebloc/.github.

Acceptance

• Selector drift (docs -l app= ↔ chart pod labels) caught in CI
• Sizing drift (preflight thresholds ↔ docs sizing numbers) caught in CI

Refs #218, #217, tracebloc/backend#746