Feat/copilot cli runner

[danymarques]

Integrate @github/copilot as a new code generation runner. The runner
supports multiple models (Claude, Gemini, GPT), configures permissions
via .copilot/settings.json, and uses COPILOT.md for instructions.

Also, add debug logging to BaseCliAgentRunner, controllable via the
CLI_RUNNER_DEBUG environment variable, to aid in troubleshooting
agent process execution.

[devversion]

Can you please just have the Copilot commit in here for easier review?

[devversion]

From talking with one of my colleagues, it sounds like we might want to not get into ownership of all sorts of runners; so maybe it would be better to discuss options in #254?

It should be possible to wire this up with the custom executor API for Web Codegen Scorer (WCS). I think we maybe want to have some good docs / examples on this. Feel free to help/play with this

[danymarques]

Thanks for the feedback!

I'd like to highlight a real-world scenario that I think is quite common: in many enterprises — including the one I work for — developers only have access to GitHub Copilot and do not hold personal API keys. Authentication and license assignment are handled via Microsoft login (Entra ID / Azure AD), and access to the underlying model goes entirely through the Copilot interface.

This isn't an edge case — it applies to any company using Microsoft Entra ID with a Copilot license. In that setup, there is no way for a developer to provide an API key, yet they do have full access to a capable model through Copilot.

Would it be possible to consider this as a supported configuration?