feat(CP-27): dashboard projection API with Monte Carlo

[ncamaa]

Summary

• Add GET /api/plans/:planId/projection endpoint that loads all plan data from the database, transforms Prisma models into engine input types, runs the 36-year deterministic projection, and returns chart-ready arrays with a summary object
• Add GET /api/plans/:planId/projection/monte-carlo endpoint that runs 1000-iteration Monte Carlo simulation with randomized return rates (Box-Muller normal distribution) and returns success rate, median final value, and p5/p25/p50/p75/p95 percentile bands
• Implement runMonteCarloSimulation() in the engine layer (src/engine/monteCarlo.ts) - pure function, no DB access
• Handle incomplete/empty plan data gracefully (valid minimal projection with zeros instead of errors)
• Register both endpoints in OpenAPI/Swagger documentation

New files

• src/engine/monteCarlo.ts - Monte Carlo simulation engine
• src/api/projection/projectionModel.ts - Zod schemas for request/response
• src/api/projection/projectionService.ts - Data loading, Prisma-to-engine transformation, reference data loading
• src/api/projection/projectionController.ts - Request handlers
• src/api/projection/projectionRouter.ts - Express routes + OpenAPI registry
• src/engine/__tests__/monteCarlo.test.ts - MC engine unit tests
• src/api/projection/__tests__/projectionService.test.ts - Projection data transformation + edge case tests

Modified files

• src/engine/index.ts - Export runMonteCarloSimulation
• src/server.ts - Mount projection router
• src/api-docs/openAPIDocumentGenerator.ts - Register projection registry

Test plan

• 17 new unit tests covering empty plans, minimal plans, edge cases, and Monte Carlo output structure
• All 228 tests pass
• pnpm build succeeds (prisma generate + tsc + tsup)
• Biome lint/format clean
• cspell spelling clean
• curl test against running dev server with sample plan data
• Verify Swagger UI shows new Projection endpoints