chains/benchmarks

[dovinmu]

Goal for this branch: demonstrate at least a few benchmarks implemented in AIPL, starting with BIG Bench

• [ ] generalize the llm op to route to different specific LLM sources
• [ ] write out metrics to JSON file
• [ ] generic bigbench binary classification script

[dovinmu]

closing in favor of #30