docs: choosing a method — baseline vs structured

[Frankie-Xu]

Summary

• Add docs/en/concepts/choosing-a-method.md and the Chinese counterpart: when graph extraction is worth the ingest cost (multi-hop, relationship-dense queries) vs when chunk_rag is enough (lookup, original wording, zero LLM ingest).
• Note that chunk_rag and graph_rag share provenance, so a same-corpus comparison is apples-to-apples. Explicitly not an eval harness.
• Hang the page in mkdocs.yml immediately after concepts/methods.md. Link from methods.md and using-methods.md (EN + ZH).

Related to #117

Out of scope

• Eval harness / retrieval metrics / QA pairs (item (3) on RFC: chunk_rag as method-baseline eval vs graph_rag #117)
• Changing method implementations
• Closing RFC: chunk_rag as method-baseline eval vs graph_rag #117 (the comparison script is the remaining (2))

Test plan

• Nav entries sit after methods.md in both EN and ZH
• Cross-links from the existing methods pages