WIP: Use playwright actions api

[craigmulligan]

Fixes #55.

TODO:

• Pull in the javascript code generator from playwright and use that instead of manually generating code.
• Test that the prompt doesn't in fact need the full schema hardcoded.

[zachblume]

Awesome start!

What are your thoughts as we dip into modularization -- my feeling is that it may end up making more sense to modularize and co-locate code by phase (discovery/execution/caching/accumulation of tests) rather than by e.g. e.g. a single schema.ts where we declare all the schema types for all the modules. I was also thinking this through re: prompts as well. They are a bit different because you might imagine people wanting to tweak the prompts more frequently, across all modules. SWE-agent took the approach of placing everything configurable AND prompts into a single config file, e.g. https://github.com/princeton-nlp/SWE-agent/blob/main/config/default.yaml

[craigmulligan]

Yeah, I really like organizing it by phase, that makes a lot of sense. I'll move schemas back into index.ts and then we can tackle re-organizing things later.