Agent benchmarks

[dimavrem22]

Resolved:

• Agent Performance Test Suite #68
• URL of DownloadOperation not interpolated by session/local storage and cookies #73
• sessionStorage, localStorage, and cookies are not resolved by the JS if they are in the keys of dict! #72

Benchmark Flow:

• python web_hacker/scripts/run_benchmarks.py -v
• has a list of s3 benchmarks fixtures
• downloads one benchmark test at a time and unzips it (contains cdp captures, routine discovery task, deterministic tests, and llm tests)
• runs discovery agent, then deterministic tests, then llm tests to produce an output like this