Add code generation analysis

[zorazrw]

Analysis of Code Generation

This PR adds an analysis for code generation, supporting:

• two datasets: HumanEval and ODEX
• execution-based evaluation using test cases
• models from HuggingFace, OpenAI, Cohere, etc.

Blocked by

• NA

[neubig]

OK, after much poking around this is working for the basic models! Here are some scores on odex for codegen and gpt2:

I will merge this branch in and we can continuously improve in future PRs. Thanks for the contribution @zorazrw !