feat(observability): k6 load test suite for critical paths

[chitrank2050]

Problem

No performance baseline exists. We don't know how many requests/second the system handles, where it bottlenecks, or what happens under load.

Proposed solution

Create k6 test scripts for critical paths:

1. Auth flow: login → use token → refresh → logout (50 VUs, 5 minutes)
2. Usage ingestion: batch of 10 events per request (100 VUs, 5 minutes)
3. Entitlement check: GET /entitlements/api_calls/check (200 VUs, 5 minutes — this is the hot path)
4. Invoice generation: generate + finalize (10 VUs, 2 minutes)
5. Mixed workload: realistic traffic distribution across all endpoints

Each test records: requests/sec, latency p50/p95/p99, error rate, throughput.

API design (if applicable)

N/A — k6 test scripts in tests/load/ directory.

Alternatives considered

• Artillery — acceptable alternative. k6 is more widely used and has better Grafana integration.
• JMeter — rejected. XML configuration is painful. k6 uses JavaScript.

Additional context

Tests run against Docker Compose stack. Results exported to Grafana for visualization.