feat: per-activity and per-workflow token use tracking and cost estimation (#232)

[m2ux]

Summary

Adds per-activity and per-workflow token use tracking and cost estimation to the workflow-server, persisting rolled-up usage in session.json with optional harness-relayed cost estimates from a versioned price table.

🐛 Issue 📐 Engineering 📋 Test plan

---

Motivation

Workflow sessions provide no visibility into the token consumption or monetary cost of executing a workflow. Token use cannot be attributed to individual activities, so operators cannot identify expensive activities, compare workflow efficiency across runs, or account for what a completed work package cost to execute.

Without a durable record, that information is lost the moment a session ends. Tracking usage per activity, aggregating it per workflow, and logging the results to the planning artifact gives every work package a permanent usage and cost record.

---

Changes

• Server — history event — Added usage_recorded to the shared history-event vocabulary for per-activity audit events.
• Server — session schema — Added optional usage field to SessionFile (per-activity record + child-inclusive workflowTotal), absent-until-first-use.
• Server — serialization — Registered usage in TOP_LEVEL_KEY_PRIORITY for canonical session ordering.
• Server — cost estimation — Added versioned config price table and estimateCost helper with derived cache multipliers; unknown model yields cost_usd null.
• Server — roll-up — Added sumUsageTree for completion-time child-inclusive workflow totals via in-tree traversal.
• Server — capture seam — Wired optional usage param on next_activity; record at activity exit, roll up incrementally, fold children at completion.
• Corpus — populate instruction — Direct the orchestrator to relay harness-reported completion usage on each next_activity call (graceful degradation when unavailable) — corpus tasks 7–8 tracked separately in workflows worktree.
• Corpus — completion renderer — Technique to render NN-token-usage.md + README summary/link from rolled-up session state on completion — corpus tasks 7–8 tracked separately.

Tests — 17 cases across unit, integration, and e2e scopes mapping to SC-1..SC-6: schema round-trip, cost estimation, child roll-up, next_activity capture, graceful omission, and completion artifact rendering.

Key decisions: Declared usage param + config price table (not OTEL ingest or server-side re-tokenization); server-first build sequence; cost labelled as estimate with subscription caveat; server owns structured state, corpus owns human-readable artifacts.

---

🤖 AI Assistance

• Assistant / Model: Cursor / composer-2.5-fast
• Context scope: mixed
• Prompt classes: code-generation, test-writing, refactoring, docs
• Provenance log: not recorded — task rows were not appended during implement

---

📌 Submission Checklist

• Changes are backward-compatible (or flagged if breaking)
• Pull request description explains why the change is needed
• Self-reviewed the diff
• I have included a change file, or skipped for this reason: repo has no changes/ folder convention
• If the changes introduce a new feature, I have bumped the node minor version — N/A (Node MCP server, not node runtime)
• Update documentation (if relevant)
• No new todos introduced

---

🔱 Fork Strategy

• Node Runtime Update
• Node Client Update
• Other
• N/A

---

🗹 TODO before merging

• Ready for review