feat(03-01): backend API endpoints — channels, billing, usage, and audit logger enhancement

- Create channels.py: HMAC-signed OAuth state generation/verification, Slack OAuth install/callback, WhatsApp manual connect, test message endpoint
- Create billing.py: Stripe Checkout session, billing portal session, webhook handler with idempotency (StripeEvent table), subscription lifecycle management
- Update usage.py: add _aggregate_rows_by_agent and _aggregate_rows_by_provider helpers (unit-testable without DB), complete usage endpoints
- Fix audit.py: rename 'metadata' attribute to 'event_metadata' (SQLAlchemy 2.0 DeclarativeBase reserves 'metadata')
- Enhance runner.py: audit log now includes prompt_tokens, completion_tokens, total_tokens, cost_usd, provider in LLM call metadata
- Update api/__init__.py to export all new routers
- All 27 unit tests passing

packages/orchestrator/orchestrator/agents/runner.py

@@ -182,6 +182,19 @@ async def run_agent(
             input_summary = _get_last_user_message(loop_messages)
             output_summary = response_content or f"[{len(response_tool_calls)} tool calls]"
             try:
+                # Extract token usage from LLM pool response
+                usage_data = data.get("usage", {}) or {}
+                prompt_tokens = int(usage_data.get("prompt_tokens", 0))
+                completion_tokens = int(usage_data.get("completion_tokens", 0))
+                total_tokens = prompt_tokens + completion_tokens
+
+                # Extract cost from response or estimate from usage
+                cost_usd = float(data.get("cost_usd", 0.0))
+
+                # Extract provider from model string (e.g. "anthropic/claude-sonnet-4" → "anthropic")
+                model_str: str = data.get("model", agent.model_preference) or ""
+                provider = model_str.split("/")[0] if "/" in model_str else model_str
+
                 await audit_logger.log_llm_call(
                     tenant_id=tenant_id,
                     agent_id=agent_uuid,
@@ -190,9 +203,14 @@ async def run_agent(
                     output_summary=output_summary,
                     latency_ms=call_latency_ms,
                     metadata={
-                        "model": data.get("model", agent.model_preference),
+                        "model": model_str,
+                        "provider": provider,
                         "iteration": iteration,
                         "tool_calls_count": len(response_tool_calls),
+                        "prompt_tokens": prompt_tokens,
+                        "completion_tokens": completion_tokens,
+                        "total_tokens": total_tokens,
+                        "cost_usd": cost_usd,
                     },
                 )
             except Exception: