feat: add bounded AgentsKit tool-calling example

[EmersonBraun]

What changed

This adds a self-contained AgentsKit workspace to the TypeScript examples. It uses OpenRouter as the model gateway and a bounded AgentsKit runtime to investigate a synthetic checkout incident with two local tools.

The example preserves tool-call identity, limits execution to four model steps, returns structured tool results, and prints an inspectable run summary. OPENROUTER_MODEL selects the model without changing the tools, orchestration, or application contract.

Why

Calling a model is only one part of an agent. The surrounding loop also needs predictable stopping behavior, tool-result correlation, and enough execution data to understand what happened.

This example keeps that control boundary small and provider-independent while still using OpenRouter's unified model gateway. The synthetic tools can later be replaced with metrics or deployment clients without rewriting the model integration.

Validation

• bun --filter '@openrouter-examples/agentskit' typecheck
• bun --filter '@openrouter-examples/agentskit' test — 3 deterministic tests, 11 assertions
• bunx biome check agentskit
• bun audit --audit-level=high
• git diff --check
• One live run with openai/gpt-oss-20b:free: 3/4 steps, both local tools completed, tool-call IDs remained correlated, and all required answer sections were present

The automated tests are credential-free and make no external requests. No key, raw provider payload or account data is included in the contribution.