docs(i18n): sync minority-language docs (de/es/fr/ko/zh-cn) to current ja source (15.7 + 15.8)

[marevol]

Summary

Synchronizes the minority-language translations (de, es, fr, ko, zh-cn) with the current Japanese source (ja/, the source of truth) for both the 15.7 and 15.8 documentation trees. Many ja pages were corrected and expanded against the actual Fess implementation during the June–July 2026 "correct against implementation" work; those corrections were not fully propagated to the minority languages, so their translations had drifted and in places described older or incorrect behavior.

One reviewed commit per language. Docs-only, additive/corrective, and every code block, property name, value, and :doc: target is kept byte-identical to ja.

What changed (by class of fix)

• Missing content restored — whole sections and fields that ja had gained but the translation lacked: the Slack data-store guide's file-crawling / troubleshooting / advanced-script sections (es), the S3/GCS crawler-configuration section and the s3,gcs entry in crawler.file.protocols (ko), dropped JSON response fields (scheduler id/created), and dropped field qualifiers.
• Mistranslations / leftover source text — a nonsensical rendering of "production environment" in sso-entraid (ko), leftover Japanese fragments inside prose and code examples across languages (localized to the target language; verified zero leftover kana in all 139 files).
• Code-example fidelity — doubled backslashes in Groovy scripts (\\n → \n), reordered system.properties / fess_config.properties blocks to match ja, and reverted a handful of over-localized non-Japanese example values (your-fess-domain.com, my-gcp-project, <strong_password>) back to ja.
• Build-warning fixes — heading underlines shorter than the title were extended (for ko/zh-cn this uses display width, since docutils' column_width counts CJK glyphs as 2); one malformed :doc: role that would fail the Sphinx build was repaired.
• Orthography — restored stripped diacritics in a number of de/fr files that had been reduced to plain ASCII (see the note below on the remaining scope), and repaired ASCII-stripped umlauts in two German files (ds-json, llm-openai).
• Consistency — the elevate-word "reading" example (a Japanese phonetic-reading field) is normalized to an empty string "" in every language, matching how en renders it; each language had previously left a different incorrect value there.

How it was verified (no full Sphinx build)

Per-language, both versions, a script checked every changed file for: leftover Japanese kana, :doc:/:ref: target parity with ja, heading-underline adequacy (CJK display-width aware), and inline-literal parity with ja. All 10 language×version sets report 0 errors; the whole branch has 0 files with leftover kana. Most drift-listed files turned out to already be at parity (git-date staleness over-counts), so edits are concentrated on the files with genuine divergence.

Not included / suggested follow-up

• Diacritics normalization. Spanish and (to a lesser extent) French have a pre-existing, systemic issue where many files were translated in plain ASCII without accents — roughly ~45 es and ~45 fr files per version contain no diacritics at all. This PR restored accents only in files it was already editing for parity; a dedicated copy-editing pass would be the right way to normalize the rest consistently. German is unaffected (umlauts are intact apart from the two files fixed here).
• A few pre-existing, out-of-scope style inconsistencies were left untouched to minimize churn: fullwidth vs halfwidth punctuation in some ko files, and a malformed :doc: role that also exists in en/15.8/install/troubleshooting.rst (fixed on the translation side only).

139 files across de, es, fr, ko, zh-cn (15.7 + 15.8).