Add contribution guidelines and issue/PR templates

[alexreinking]

Summary

• Add CONTRIBUTING.md covering the issue-first workflow, testing expectations, CI requirements, AI attribution policy, and review/merge process.
• Add GitHub issue form templates for bug reports, missed optimizations, and feature specifications.
• Add issue template config that redirects Q&A and brainstorming to Discussions.
• Add a PR template with a performance impact section and checklist for tests and AI attribution.

Test plan

These can only be verified after merging, since GitHub renders issue/PR templates from the default branch:

• Verify issue templates render correctly: go to Issues > New Issue and check all three templates plus the Discussion links.
• Verify PR template renders correctly: open a draft PR and check the pre-filled body.
• Verify CONTRIBUTING.md renders correctly on GitHub and links resolve.
• Review wording with other maintainers.

This is the result of a back-and-forth drafting process with Claude Code.

[mcourteaux]

I think we should maintain a list of the exact co authored by tags for each LLM (and in particular which email address to use.)

[mcourteaux]

The referenced code of conduct does not mention LLM attribution?

[alexreinking]

The referenced code of conduct does not mention LLM attribution?

It does, though? Under "Give credit where it's due."

Additionally, any code generated by an AI tool (such as a large language model) must be clearly identified as such in the commit message using "Co-authored-by: The name of the tool". We all like recognition for our work.

[mcourteaux]

Oops. My bad. Will approve if there is a list of LLM tags.

[alexreinking]

Oops. My bad. Will approve if there is a list of LLM tags.

There is no real consensus on how coding agent tools include attribution or not. Here's the conventions I've found so far, from most structured to least:

• Claude uses Co-authored-by: Claude $MODEL <noreply@anthropic.com>
• Cursor uses Co-authored-by: Cursor <cursoragent@cursor.com>
• GitHub Copilot authors commits as a bot account. When it isn't the primary committer, we could insist on Co-authored-by: GitHub Copilot <$GITHUB_USER_ID+Copilot@github.com>
• Codex (OpenAI) has no standard process. If you ask it, it tells you to use Generated-by or Assisted-by, neither of which are standard in Git's tooling. Another time I asked it, it suggested to use Co-authored-by: Codex <noreply@openai.com>. Using Co-authored-by: chatgpt-codex-connector[bot] <199175422+chatgpt-codex-connector[bot]@users.noreply.github.com> is plausible, too, see the discussion on openai/codex.
• Gemini CLI - Also nothing standard. Discussion on google-gemini/gemini-cli suggests Co-Authored-By: Gemini 3 <gemini-code-assist@google.com> works.

All of these can be disabled or modified in their respective application settings. We'll need to discuss what to ask of contributors and how we as maintainers need to enforce at merge time.