The `always_allow_tool_actions` setting would get overridden with the
default when we loaded each example project, leading to examples
stalling when they run a tool that needed confirmation. There's now a
separate `runner_settings.json` file where we can configure the
environment for the eval.

Release Notes:

- N/A

---------

Co-authored-by: Oleksiy <oleksiy@zed.dev>