feat(#233): extract scheduler engine into pure lib (Phase 2 of Resource Optimiser)

[NickMonrad]

Summary

Phase 2 of 4 for Resource Optimiser (#233). Extracts the timeline scheduling engine from the Express route handler into a pure, testable library (server/src/lib/scheduler.ts) so the Phase 3 optimiser can call it in a tight loop with different resource configurations — no DB round-trips per iteration.

Behaviour-preserving refactor. All existing tests pass without modification. The HTTP API response shape is unchanged.

Changes

server/src/lib/scheduler.ts (NEW, 799 lines)

• Pure runScheduler(input: SchedulerInput): SchedulerOutput — no Prisma, no I/O, no module-level state
• All scheduling logic moved here: topological sort (Kahn's), MinHeap priority queue, resource-levelling simulation, story proportional scheduling, cross-epic anti-cycle guard, named-resource start/end week constraints
• Exposed types: SchedulerInput, SchedulerOutput, SchedulerEpic, SchedulerFeature, SchedulerStory, SchedulerTask, SchedulerResourceType, SchedulerNamedResource, ParallelWarning
• Helpers also exported: effectiveAllocationPct, getWeeklyCapacity, computeParallelWarnings

server/src/routes/timeline.ts (−730 lines net)

• POST /schedule now: load Prisma data → map to SchedulerInput → call runScheduler → write results to DB → respond
• Route shrank from ~1240 lines to ~510 lines
• getWeeklyCapacity re-exported for backwards compat with existing timeline.test.ts

server/src/test/scheduler.test.ts (NEW, 22 tests)

Covers happy path, multi-feature parallelism, epic dependencies, feature dependencies, resource constraints, named-resource start/end weeks, manual story overrides, empty input, zero-hour features, cross-epic dependency anti-cycle (hasCrossEpicDep guard), epic timelineStartWeek anchor.

Review fixes applied (sub-agent review cycle)

• 🟡 Perf — epics.find() in epic-dep loop replaced with Map<id, epic> lookup (O(n×d) → O(d))
• 🟡 Perf — adjList changed from Map<string, string[]> with .includes() dedup to Map<string, Set<string>> with .has() (O(e²) → O(e))
• 🟢 Coverage — added 2 missing test cases (cross-epic anti-cycle, timelineStartWeek anchor)
• ✅ Verified: computeParallelWarnings async→sync conversion is safe (had no awaits), manual upsert guards preserved, MinHeap actually used, zero non-determinism in pure path

Tests

• npx tsc --noEmit (server) — ✅ clean
• npx tsc --noEmit (client) — ✅ clean
• npm test (server) — ✅ 164/164 passing (142 existing + 22 new scheduler tests)

Phases

• Phase 1 — Snapshot expansion (feat(#233): expand snapshot JSON to capture timeline state (Phase 1 of Resource Optimiser) #234, merged)
• Phase 2 — Extract scheduler engine (this PR)
• Phase 3 — POST /projects/:id/optimise endpoint with modes (Speed / Utilisation / Balanced)
• Phase 4 — Frontend optimiser drawer + result preview/apply

Refs #233