chore(testing): extended E2E benchmark session runner

[saschabuehrle]

Adds an extended real-API testing/benchmarking session setup:

• New guide: docs/guides/extended-testing.md
• Updated benchmark README with correct commands
• tests/benchmarks/run_all.py: adds --profile presets (smoke/standard/overnight/full), optional provider comparison, and limits BFCL agentic task count
• New benchmark: tests/benchmarks/agentic_multi_agent.py (router + tool-calling correctness + cost reduction)
• One-command session runner: scripts/extended-e2e-session.sh (logs + JSON artifacts)
• TypeScript real API smoke: pnpm -C packages/core run real-api:smoke