Add llms.txt and AI-agent documentation (following tensor4all/tenferro-rs)

[shinaoka]

Summary

Following the example of tensor4all/tenferro-rs, we should provide an llms.txt and related AI-agent-facing documentation so that coding agents (Claude Code, Codex, etc.) and LLM-based tools can work with this library correctly without reading the full source.

Reference: tenferro-rs docs/llms.txt — a compact index following the llms.txt convention: one-paragraph project summary, a "conventions you must know before writing code" section highlighting silent-failure pitfalls, and curated links (README, guides, API reference, agent skill).

Proposed tasks

• Add llms.txt (e.g. under docs/, served from the documentation site root as well):
  • One-paragraph description of the library (sparse IR basis, tau/Matsubara sampling, DLR)
  • Conventions you must know section, e.g.: memory layout expected at the C-API boundary, positive_only semantics (real-coefficient assumption), complex vs real expansion coefficients, tau ∈ [0, β] conventions, statistics (fermionic/bosonic) sign conventions, accuracy vs eps of the basis
  • Curated links: README, tutorials, API docs
• Add AGENTS.md (agent-facing repository guide: build/test commands, repo layout, contribution rules) — keep CLAUDE.md as a thin pointer to it if both exist
• (Optional) an agent skill (.agents/skills/ or .claude/skills/) helping downstream agents choose the right API (sampling vs DLR, fit/evaluate axis conventions) and avoid known pitfalls
• Link llms.txt from the README so it is discoverable

Notes

• The same effort is tracked in the sister repositories (Rust / Julia / Python) so the three stay consistent; shared wording for the physics conventions can be drafted once and reused.

[shinaoka]

Sister issues: SpM-lab/sparse-ir-rs#258, SpM-lab/SparseIR.jl#152