Summary

Add a Node node:sqlite persistence skeleton so we can test whether the
TypeScript port hits fatal language-switch pits (Analyzer v1, INTEGER,
FTS projection, CAS, and refusing unstamped databases).
This is an experimental self-created-database path only. It does not
implement the shared-schema gate in #3, does not write existing Python
databases, and does not claim sqlite-fts / C3.

Profile and parity

• Profile: sqlite-fts (experimental skeleton only; not a product profile claim)
• Target level: C0 in-process kernel + Analyzer C2 samples vs pinned Python
• Related issue or RFC: relates to SQLite kernel and schema-version gate #3 / Memory, FTS, and PreparedContext #4; does not close them. ADR 0002
  (experimental DB allowed; no shared-write / C3 claim). Analyzer identity
  powercontext.analyzer.v1.

Checklist

• Does not add a second wire contract
• Does not hand-edit contract/ or conformance/ snapshots
• Does not hand-edit packages/protocol/src/generated
• Baseline bump, if any, is this pull request’s only purpose (N/A)
• New or changed behavior names the compatibility decision or ADR
  (ADR 0002 experimental DB; SHARED_DATABASE_WRITES_ALLOWED = false)

Test plan

pnpm format
pnpm lint
pnpm typecheck
pnpm test
pnpm generate:check

Optional Analyzer oracle (pinned oceanbase/powercontext at 733e4bf6b378785e76274ff07632029c699ecb09, not PowerMem):

POWERCONTEXT_SKELETON_ORACLE=1 POWERCONTEXT_PYTHON_ROOT=<real-powercontext> \
  pnpm test packages/core/tests/analyzer-oracle.test.ts

Out of scope

• Closing SQLite kernel and schema-version gate #3 (still blocked on Python ADR 0001 / 0002 + baseline bump)
• Offline Python ↔ TypeScript reads of existing v0.0.2 databases
• HTTP Server / MCP / CLI (Subset Server and cross-language HTTP E2E #5)
• Memory revise / retire / PreparedContext C3 (Memory, FTS, and PreparedContext #4)