docs(locale): translate all remaining chapters into Japanese

[sanak]

Overview

Completes the Japanese translation of the workshop. This PR translates the catalogs
placed by PR #1, bringing every message catalog to 0 untranslated and 0 fuzzy.

Base branch is locale-ja (PR #1), not develop, so that this PR shows only the
translation work and not the catalog placement from PR #1.

Changes

Translated 1,171 units across 25 catalogs. basic/graphs (163 entries) and the
14 fuzzy entries were already completed in PR #1, so all 1,461 units of the workshop
are now translated.

Category	Catalogs	Units translated here
basic	5	418
un_sdg	5	433
general-intro	3	94
interactions	3	67
appendix	4	85
advanced	1	46
examples	3	22
index	1	6
Total	25	1,171

Translation policy

Follows tmp/glossary-ja.md. Every rule was chosen to stay consistent with the
translations the workshop already had.

• routing is translated as ルート検索
• Trailing long-vowel marks are dropped from katakana loanwords (レイヤ, サーバ,
  パラメータ, ユーザ, ベクタ, フィーチャ, クエリ), with トポロジー, メモリー, データ
  and キー as established exceptions
• UN SDG goal names use the official Japanese wording published by MOFA and the
  UN Information Centre
• GUI labels are kept in English — the contents of :menuselection:, *Login*,
  GeoServer connection parameter keys, and similar strings that a reader has to
  match against what is on screen
• pgRouting function names, table names, column names and product names are not translated
• column / row / field are translated as 列 / 行 / フィールド
• The same msgid gets the same translation across catalogs. This applies to prose,
  UI strings and instructions; for section headings the surrounding chapter's style
  takes precedence, which accounts for the single deliberate exception below

Verification

• 0 untranslated and 0 fuzzy in all 26 catalogs — msgfmt --statistics reports
  1,461 translated / 0 untranslated / 0 fuzzy
• msgfmt -c passes for all 26 catalogs
• check_terms.py reports 0 terminology violations
• check_rst.py reports 0 broken msgstr — it parses each msgstr with docutils,
  which catches markup that silently breaks without failing msgfmt
• check_spacing.py reports 0 violations — missing spaces between Japanese and
  Latin text
• Cross-catalog scan finds only one msgid translated two different ways, and
  that one is the deliberate exception documented below
• No msgid was modified — msgid sets compared against the locale-ja versions
  with polib; all 25 files identical
• make html-ja completes with 0 Sphinx warnings, from a clean doctrees
• All 25 pages match en structurally — heading levels, internal and external
  references, images, <cite>, <code>, tables and system messages all have
  identical counts
• All 25 pages render their <h1> in Japanese; no page is left in English
• Visual check with Playwright on 7 pages — layout, table overflow, images,
  Japanese font rendering, and no Japanese leaking into code blocks
• Only locale/ja/ is changed (25 files); no .mo files
• No drift in locale/pot against upstream

Deliberate exception: Prepare the database

This msgid is translated two different ways, and both are intentional:

• basic/data.po → データベースの準備, matching that chapter's noun-phrase headings
• un_sdg/sdg11-cities.po → データベースを準備する, matching that chapter's verb-form headings

Both occurrences are section headings, and each chapter is internally consistent.
Any additional cross-catalog divergence would be a regression.

Known issues (not addressed in this PR)

These behave identically in the en build and originate in the English source or in
Sphinx, not in the translation.

• Substitutions are not expanded inside an image :alt: value, so |place_1| and
  |place_3| appear literally in basic/vehicle.html, and |ch7_place_1| /
  |ch7_place_2| in basic/sql_function.html. docutils does not resolve
  substitution references in option values. Same count and location in en.
• check_terms.py reports 2 markup violations and check_rst.py reports 4 warnings.
  Both are structural false positives caused by the English source: a stray leading
  backtick in general-intro/overview, and msgids in osgeolive that end in ::
  while the literal block itself lives in a separate msgid. The rendered HTML was
  verified to be correct in both cases.
• Sphinx smartquotes behaves differently per language, so ... renders as … in en
  but as three ASCII periods in ja, and "..." renders with curly quotes in en
  but straight quotes in ja.
• The language bar is not generated at all. INSERT_INTO_MAP in CMakeLists.txt
  defines LangMap_<code> variables that nothing in the repository reads. Unrelated
  to this PR, but worth noting since it is the documented way to surface 日本語.

Defects found in the English source

Roughly 20 defects in the English text were recorded while translating — missing verbs
and articles, clauses joined without conjunctions, nodded for noded, your for
you, a Do ...? question that should be Are ...?, and a hanoslav / hansolav
spelling mismatch between prose and URL. They are deliberately not fixed here, since
changing the English source would change the msgids. They are worth a separate
upstream issue.

Open item carried over from PR #1

PR #1 deferred a decision to this PR: 16 catalogs still carry sphinx-intl's default
PO-Revision-Date and Last-Translator values, and Project-Id-Version is split
across workshop generations (14 Belém 3.0 / 9 Hiroshima 3.2 / 3 Auckland 3.1).

They are left untouched here. These headers are Weblate-managed and get rewritten on
sync, and normalizing them would add churn across 16 files that is unrelated to the
translation itself. Happy to do it in a follow-up if preferred.

Out of scope

• Submitting to upstream pgRouting/workshop
• Propagating to Weblate

[sanak]

Close this PR, because upstream merged the following PR.
pgRouting#284