B2B-15 · Copilot metering: usage log, token accounting, quotas

Part of the B2B epic (Bloco 4 · item 4.1). Today there is **zero metering/quota/token accounting** on any LLM call (`ai-task/` queue, `copilot/copilot-chat.service.ts` — GPT-5-mini via LangChain).

## What
- [ ] `UsageLog` collection (userId / org / namespace / feature / tokens / cost)
- [ ] Capture `usage` from LangChain calls (`copilot-chat.service.ts:317`, `ai-task.service.ts:69-78`) — instrument **every** LLM call site
- [ ] Quota enforcement
- [ ] Aggregation per seat/org + billing hook

**Effort:** ~2–3 SP · **Depends on:** B2B-5, B2B-8

Start by reading the LLM call sites named in copilot/copilot-chat.service.ts:317 and ai-task.service.ts:69-78, then map the remaining call sites and review the B2B-5 and B2B-8 dependencies. Done means usage is captured for every LLM call, UsageLog records exist, quotas are enforced, and seat/org aggregation plus the billing hook are covered.

索引モデルが issue の本文から書いたものです。