feat(core-realtime): scaffold realtime optional core

Generator-emitted scaffold (pnpm turbo gen core-package realtime) plus
the story-00-precedent coverage repairs (coverage provider devDep,
symbols.ts exclude + tested allowlist mirror) and three minimal tests
covering generator-emitted realtime code the template suite misses.
Squash of 31d85e0 + review-fix cf11b38.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_016j8z4VHjedXDTjEDNg7qHK

packages/core-realtime/docs/library-decisions/2026-05-14-socket.io.md

@@ -0,0 +1,72 @@
+---
+package: socket.io
+version: "^4.7.0"
+tier: core
+decision: approved
+date: 2026-05-14
+deciders: [scaffolded]
+adr: adr-016
+filter-results:
+  license: MIT
+  types: native
+  maintenance: active
+  boundary-fit: pass
+  shadow-check: pass
+  eu-residency: self-hostable
+  cve-scan: clean
+  named-consumer: pass
+verification-commands:
+  - pnpm audit --audit-level=moderate
+  - npm view socket.io license
+accepted-cves: []
+---
+
+## Filter: license
+
+MIT — on the workspace allowlist.
+
+## Filter: types
+
+Ships first-party TypeScript types in its distribution.
+
+## Filter: maintenance
+
+Active. Maintained by the Socket.IO team; frequent releases and active issue tracker.
+
+## Filter: maintenance
+
+Active. Regular releases; widely deployed in production.
+
+## Filter: boundary-fit
+
+ADR-016 §R2 explicitly designates `core-realtime` as the sole allowed home for `socket.io`. Boundary rule `no-direct-socket-io` enforces this in ESLint.
+
+## Filter: shadow-check
+
+No competing realtime transport in the workspace. No shadow.
+
+## Filter: eu-residency
+
+Self-hosted server; the library itself does not transmit data to any vendor endpoint.
+
+## Filter: cve-scan
+
+No advisories at adoption time.
+
+## Filter: named-consumer
+
+`core-realtime` wraps socket.io to provide the `IRealtimeServer` abstraction (ADR-016).
+
+## Prompt: replaces
+
+Nothing — this is the initial realtime scaffolding. No prior transport to retire.
+
+## Prompt: migration-cost-out
+
+Hard: channel descriptors, handler signatures, and server-side broadcast API are all shaped around socket.io semantics. Replacing requires re-implementing the abstraction layer.
+
+## Prompt: alternatives-considered
+
+1. **ws** — lower-level, no rooms or namespaces; would require significant protocol work.
+2. **Ably / Pusher** — vendor-hosted; eu-residency risk and ongoing cost.
+   Socket.IO is the established standard for this use-case and is fully self-hostable.

packages/core-realtime/docs/library-decisions/2026-05-14-zod.md

@@ -0,0 +1,66 @@
+---
+package: zod
+version: "^3.23.0"
+tier: core
+decision: approved
+date: 2026-05-14
+deciders: [scaffolded]
+adr: adr-016
+filter-results:
+  license: MIT
+  types: native
+  maintenance: active
+  boundary-fit: pass
+  shadow-check: pass
+  eu-residency: n/a
+  cve-scan: clean
+  named-consumer: pass
+verification-commands:
+  - pnpm audit --audit-level=moderate
+  - npm view zod license
+accepted-cves: []
+---
+
+## Filter: license
+
+MIT — on the workspace allowlist.
+
+## Filter: types
+
+Ships first-party TypeScript types in its distribution (`.d.ts` included).
+
+## Filter: maintenance
+
+Active. Regular releases by Colin McDonnell; widely adopted.
+
+## Filter: boundary-fit
+
+Core package. Zod is the workspace-canonical validation library locked in `core-shared` (ADR-016).
+
+## Filter: shadow-check
+
+Zod is already the workspace-locked validation library. No shadow.
+
+## Filter: eu-residency
+
+Pure computation; no network calls or vendor data transmission. n/a.
+
+## Filter: cve-scan
+
+No advisories at adoption time.
+
+## Filter: named-consumer
+
+`core-realtime` uses zod for channel descriptor and payload schema validation.
+
+## Prompt: replaces
+
+Nothing — zod is the pre-existing workspace validation library.
+
+## Prompt: migration-cost-out
+
+Mechanical: swap schema definitions at call sites. No data-format lock-in.
+
+## Prompt: alternatives-considered
+
+Zod is workspace-locked (see `core-shared`). A replacement would require a workspace-wide ADR; no alternative was evaluated here.