feat: add api benchmark suite with regression tracking

[KarelTestSpecial]

Description

This PR adds a comprehensive API benchmarking suite to the project as requested in issue #30.

Key Features

• Benchmarking Tool: Integrated autocannon for high-performance load testing.
• Reporting: Automated generation of JSON and Markdown reports in /benchmarks/results/.
• Regression Tracking: Configurable thresholds in thresholds.json to detect performance regressions.
• CI Integration: GitHub Action workflow that runs a smoke benchmark on every PR and fails if thresholds are exceeded.
• Single Command: Run the full suite with npm run benchmark (or pnpm benchmark) from the benchmarks directory.

Benchmark Environment Disclosure (Required)

Hardware

• CPU model & core count: Intel(R) Core(TM) i5-8265U CPU @ 1.60GHz (8 threads)
• RAM (total & available during benchmark): 32GB (28GB available)
• Storage type (SSD / NVMe / HDD): SSD (ChromeOS Virtual Disk)
• Network interface (Ethernet / WiFi / loopback): loopback (local staging)
• Machine type: local workstation (Chromebook Linux Container)
• OS & version: Debian 12 (Bookworm) Kernel 6.6.99

Runtime

• Node.js version: v22.22.0
• Any resource limits applied: No strict cgroup limits.
• Other significant processes: None during benchmark.

AI Agent Disclosure

• Agent: Gemini CLI (Custom Agentic Workspace)
• Model: Gemini 3.1 Pro
• Execution: Human-supervised (Karel) / Autonomous exploration
• Shell/Tool Access: Yes
• Internet Access: Yes
• Benchmark Commands: Run by agent directly (supervised)

Closes #30

[KarelTestSpecial]

Closing this PR. After further review, this project does not meet our contribution criteria.