XY-911 Optimize Decodex operator status reads

[yvette-carlisle]

Summary

• Add lazy runtime-state opening for operator status so local snapshots do not hydrate the full protocol event journal.
• Add project-scoped metadata/evidence loaders and SQLite point lookups for issue attempts, worktrees, contracts, execution programs, and connector backoff.
• Keep status/dashboard loop and review projections backed by one project loop-evidence snapshot per render.
• Slim Decodex automation/manual-cli skills after plugin-eval while preserving tested research/promotion/queue boundaries.

Evidence

• cargo make check passed: 1048 tests, 1 skipped.
• cargo run -p decodex --bin decodex -- probe stdio:// passed: PROBE_OK.
• ./target/debug/decodex status --json: real 0.15, no active/queued/attention/warnings.
• ./target/debug/decodex status --live --json: real 8.49, no active/queued/attention/warnings.
• ./target/debug/decodex run --dry-run: real 6.98, no eligible issue found; no tracker mutation.
• plugin-eval: automation skill 100/A, manual-cli skill 100/A, plugin bundle 91/B with 0 fail and only aggregate token-budget warnings.

Notes

• Covers the status-state churn subset of XY-911 and related XY-914.
• This PR does not use merge queue and should land through the Decodex/admin merge path.