Document tracker contribution flow

[chrmod]

Summary

Move guidance on adding a new tracker into documentation that serves human contributors first and AI assistants by extension, so the dataset stays hand-curated rather than auto-generated.

• docs/contributing.md walks contributors through adding a tracker: repository layout, file formats, slug conventions, choosing a category, when to add filters, sourcing expectations, how to reuse an existing organization, and how to avoid duplicates.
• AGENTS.md at the repository root gives a brief orientation: what the project is, the curated-not-generated stance, the CI commands to run before opening a PR, and a pointer to the contributor guide.
• The skill is renamed from add-tracker-domain to research-tracker-domain and rescoped: it no longer writes files in db/. Instead it produces a sourced research report (existing coverage, operator, suggested category, candidate organization to reuse, open questions) that a contributor reviews and turns into a PR.
• README.md links to the new contributor guide.

This responds to review feedback that auto-generation tooling could send the wrong message about the data-quality bar — documentation is dual-use, and the helper now stays on the human's side of the loop.

Test plan

• Read docs/contributing.md end-to-end and confirm a new contributor would have what they need to add a tracker.
• Read AGENTS.md and confirm it gives a fresh contributor a clear next step.
• Click through the internal links in both files and from the README and verify they all resolve.
• Invoke the renamed skill on a sample domain and confirm the output is a research report — no files written under db/.

[philipp-classen]

It is a big topic, but I feel it is not in the best interest of the project to add tooling for auto-generating changes. It can send the wrong message, since we do not want to attract generated pull requests, but push for a high quality, trusted data set.

My own view is closer to https://github.com/ggml-org/llama.cpp/blob/master/AGENTS.md

Overall, I see more value in traditional documentation. It is dual-use, since it serves humans, but AI agents can also read through it.

[chrmod]

All good. Will adopt.