Run against SWE-Bench

[batwood001]

Feature description

Devin, the "First AI software engineer" is using their SWE-Bench performance as primary evidence of their capability.

Motivation/Application

Provide a benchmark to evaluate Devin's claims, and solidify gpt-engineer's reputation as a legitimate autonomous coding agent.

[ErikBjare]

I also had the urge to implement this after the Devin announcement 😄

But there is already an issue for it: #913

Closing as duplicate, continuing the discussion in the above issue.