feat(security): preflight audit setup and define complete baseline evidence

[leemour]

The next audit now has an executable preflight and an explicit first-baseline path. security preflight reports tools/pin age, duplicate physical checkouts including symlinks, public GitHub origins/access, CodeQL availability and optional previous-run validity without scanning source or creating run artifacts. Incomplete historical records remain unchanged and coverage-mismatched comparisons still refuse fixed-finding claims.

The runbook uses configurable public/private roots, supports cli-core's canonical symlink and the wirecat-github checkout, documents source-build invocation and an isolated pinned tool environment, and separates scanner completion from manual/Socket/A7 stages. Reviewer prompts assign CI/deploy and dependency review, take accepted-risk decisions as inputs, and link a positive/negative regression checklist and completion ledger. Private secret-scan scope and nonzero-history controls are explicit. The backlog now describes the remaining complete audit rather than a first Telegram agent run.

A shared TypeScript-config resolution issue prevented Vitest from running in an isolated worktree. Resolve the same pinned base file physically and select the local tsconfig explicitly so automatic discovery cannot climb into a different checkout. Compiler options and typechecking remain intact.

Validation: 114 tests pass, 98.43% lines / 97.58% statements / 94.72% branches / 98.09% functions; lint, types, docs checks, build and Bun smoke pass. Existing unrelated lint warnings remain unchanged. Real preflight results and operational readiness evidence belong in the private repository. No live messenger operation, credential read or package publication.