fix(mcp): report completed connected tool

[dhairyashiil]

Summary

Add an optional toolName to successful MCP execute results when a script uses exactly one distinct connected tool.

This gives MCP clients structured provenance for a completed result without changing the existing text result. Discovery and built-in calls, failed calls, and executions that use multiple distinct connected tools remain unlabeled. The internal per-call trace is not returned in the MCP response.

Review guide

• tool-invoker records a dynamic connected-tool path only after a successful call.
• engine keeps that trace internal and projects toolName only when one distinct path was used.
• Unit and MCP end-to-end tests cover the projection, dynamic dispatch, repeated calls, multiple tools, and the non-leakage boundary.

Verification

• npx bun@1.3.11 run format:check
• npx bun@1.3.11 run lint
• npx bun@1.3.11 run typecheck
• npx bun@1.3.11 run test
• npx bun@1.3.11 run --cwd e2e test:selfhost -- scenarios/tool-call-contract.test.ts

[dhairyashiil]

@RhysSullivan could you please check and let me know if this makes sense? let me know if you need any changes