test: measure scale latency and reliability

[heyjunpenn]

Closes #24

Summary

• add a reproducible scale, latency, and combined soak runner
• commit fixed-environment raw benchmark results and methodology
• distinguish achieved capacity/latency observations from an unsuccessful soak

Results

• 500 subscription nodes, 100 active listeners, 1,000 simultaneous CONNECT tunnels, and 10,000 sessions observed
• paired gateway-added CONNECT P95: 0.695 ms, excluding upstream and target network time
• 60-second combined soak: 94,368 attempts, 79,817 failures; reliability explicitly not verified, dominated by local load-generator ephemeral-port exhaustion

Validation

• node --test scripts/scale-reliability.test.mjs
• pnpm verify
• git diff --check