feat(cms): U1 content model + migration scaffold for blog/news/events

Establish content/ as the source of truth for editable content and
back-fill it from the existing detail pages, per U1 of the no-code CMS
plan.

- docs/content-model.md: front-matter schema (EN/MK/SI), body_format
  html-vs-markdown, content/events -> projects/ output mapping.
- tools/migrate_pages_to_content.py: bs4/yaml extractor; pulls title,
  subtitle, category, ISO date (from ARTICLE_DATES), hero image+caption;
  preserves the article body PLUS trailing sibling sections (event
  galleries) verbatim as body_format: html. Reports unparseable pages
  instead of emitting half-formed files.
- content/**: 51 files (17 entries x EN/MK/SI), 0 skips.
- tools/tests/test_migrate_roundtrip.py: 12 characterization tests
  (structured fields, news-has-no-subtitle, event gallery/lightbox
  preserved, Cyrillic UTF-8 round-trip, every page parses).
- .gitignore: ignore __pycache__/ and *.pyc.

U1 is unblocked and standalone. U2 (render content -> pages) is next;
U5/U6 remain blocked on Open Question Q1 (Gitea<->GitHub topology).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

docs/content-model.md

@@ -0,0 +1,97 @@
+# Content model — blog / news / events (EN / MK / SI)
+
+Source of truth for editable content. Implements **U1** of
+`docs/plans/2026-08-08-001-feat-no-code-cms-plan.md`.
+
+Today the detail pages under `en|mk|si/{blog,news,projects}/<slug>/index.html` are
+hand-authored HTML and are the source of truth. This model moves the source of
+truth into `content/`: one Markdown-with-front-matter file per **entry per
+language**. The detail HTML pages become *build output* (produced by U2).
+
+## Layout
+
+```
+content/
+  blog/<slug>/{en,mk,si}.md
+  news/<slug>/{en,mk,si}.md
+  events/<slug>/{en,mk,si}.md      # renders to the existing <lang>/projects/ output
+```
+
+- One folder per entry (the **slug**), one file per language.
+- A language file may be **absent** — that language simply is not published for
+  that entry (U2 skips it; it does not emit a broken half-page).
+- `content/events/` maps to the site's existing `projects/` output folder. The
+  URL path and the SEO date-registry key stay `projects/<slug>` for
+  compatibility (see `source_folder` below).
+
+## File shape
+
+Standard front-matter: a YAML block delimited by `---`, then the body.
+
+```markdown
+---
+type: blog                     # blog | news | event
+slug: how-to-open-slovenian-bank-account
+lang: en                       # en | mk | si
+source_folder: blog            # blog | news | projects  (URL + date-registry key)
+title: "How to open a Slovenian bank account as an international student"
+subtitle: "A practical guide for Macedonian students…"   # optional (news has none)
+category: "Student life"       # optional; blog category shown before the date
+date: "2026-08-01"             # ISO 8601 — authoritative for sorting (from ARTICLE_DATES)
+date_line: "Student life · Last checked 1 August 2026"   # verbatim human date line, optional
+hero:
+  src: "/images/blog-slovenian-bank-account.webp"        # root-relative
+  alt: "International student using mobile banking…"
+  width: "6048"
+  height: "4024"
+  caption_html: "<span class=\"fig-illustrative\">…</span> · Photo by …"  # optional
+body_format: html              # html | markdown  (see below)
+---
+<the article body, verbatim inner HTML of .article-body / .news-article-body>
+```
+
+### Field reference
+
+| Field | Required | Notes |
+|-------|----------|-------|
+| `type` | yes | `blog`, `news`, or `event`. |
+| `slug` | yes | URL slug; matches the folder name. |
+| `lang` | yes | `en` / `mk` / `si`. |
+| `source_folder` | yes | `blog` / `news` / `projects`. Drives the output URL and the `ARTICLE_DATES` key (`<source_folder>/<slug>`). |
+| `title` | yes | Plain text; rendered into `<h1>` and `<title>`. |
+| `subtitle` | no | Blog/event lead paragraph (`.article-subtitle`). News has none. |
+| `category` | no | Shown before the date on blog pages. |
+| `date` | yes | ISO date, authoritative for listing order. Sourced from `tools/seo_inject.py` `ARTICLE_DATES`, else extracted from the page. |
+| `date_line` | no | The exact human-readable date line from the page, preserved for fidelity. U2 may instead regenerate it per-language from `date`. |
+| `hero.*` | no | Lead image + caption. `src` is root-relative (`/images/…`). `caption_html` is verbatim (may contain attribution links). |
+| `body_format` | yes | See below. |
+
+### `body_format`: `html` vs `markdown`
+
+- **`html`** — the body is a verbatim HTML fragment: the page's `.article-body`
+  / `.news-article-body` element **with its wrapper**, plus any following
+  sibling sections inside the article container (notably `section.event-gallery`
+  on event pages, which sits *after* `.article-body`). This is what **migration
+  produces**, so rich elements the site already uses — `guide-box` asides,
+  `article-inline-image` figures, `event-gallery` lightbox grids, ordered lists,
+  nested links — survive **losslessly**. The U2 builder injects this fragment
+  as-is, in place of everything after the hero figure. Because migrated output
+  pages sit at the same directory depth as the originals
+  (`<lang>/<type>/<slug>/index.html`), relative asset paths inside the body
+  (`../../../images/…`) remain valid without rewriting. The blog scrollspy
+  sidebar is *not* stored — it is derived from the body's `<h2 id>` headings by
+  U2.
+- **`markdown`** — the body is Markdown. This is the intended format for **new
+  posts authored in the CMS**; the U2 builder converts it to article HTML using
+  the existing converter in `tools/build_blog.py`. The blog scrollspy sidebar is
+  regenerated from the `##` (h2) headings.
+
+Migration deliberately chooses `html` over a lossy HTML→Markdown conversion.
+The scaffold is a starting point a human proofs, **not** a guaranteed lossless
+transform — pages the extractor cannot parse cleanly are reported, not
+silently emitted.
+
+## Non-goals for U1
+
+- No CMS wiring (U7), no build/render of pages (U2), no listings/date refactor
+  (U3). U1 only defines this model and back-fills `content/` from current pages.