Better "is code" detection

Right now, "is it code" is based on a fixed set of extensions. This causes some:

**Ergonomic Issues**

It's annoying to have to specify this manually.

**False negatives**

There may be meaningful code in the repository that doesn't have one of those extensions.

**False positives**

Commonly, on real repositories a huge amount of generated files or like i18n source files get indexed. This is sort of a waste of time, and can end up returning results that aren't really what an engineer would be looking for.

In the abstract, it may be possible to have an LLM help us out with this if we prompted it in a way that provided the context of the goals of the dryad project.

No files, tests, or entry points are named. Start by locating the fixed extension set used for “is it code” detection and review how it handles generated and internationalization files; done requires an agreed approach that reduces manual configuration, false negatives, and false positives.

Written by the indexing model from the issue text.