feat: RU/EN i18n with header language switcher - #42

Replace the independently hardcoded en/ru literals in index.html's
pre-paint bootstrap script and i18n.js's SUPPORTED/DEFAULT_LANG with
a shared <meta name="i18n-languages"> tag, following the same pattern
already used for demoscene-base. Prevents the two lists from silently
diverging if a language is ever added to one and not the other.

Falls back to the current en/ru pair when the meta tag is absent
(e.g. i18n.js's vm sandbox in tests/site-smoke.test.mjs has no real
document), so existing tests are unaffected.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S8oQHBErmyHAWRedfFifZk

…r eyebrow

- Revert about.copy1 RU: 'спокойную' back to 'предсказуемую эксплуатацию'
  (engineering audience, not casual)
- Change opensource.eyebrow RU: 'Вклад в open source' → 'Принятые PR'
  (the section is about merged PRs, not open-source philosophy)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S8oQHBErmyHAWRedfFifZk

Headings and short label phrases don't take a terminal period
in either English or Russian typographic convention.

Affects: stars.title, projects.title, projects.captionStrong,
opensource.title, experience.title, about.title — both EN and RU
dictionaries, inline HTML, and Jinja templates.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S8oQHBErmyHAWRedfFifZk

.availability span styled ALL spans as green dots — both the
decorative indicator and the text span got width:7px, height:7px,
border-radius:50% and a green background. The text span collapsed
into a second green dot, leaving the label text overflowing.

Narrow the selector to .availability span:first-child so only
the empty decorative <span> becomes the dot.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S8oQHBErmyHAWRedfFifZk

…ths)

- Add data-i18n attributes to project group titles, card descriptions,
  stars section-note (split around dynamic values), SVG month labels,
  and experience timeline-org in both index.html and Jinja templates
- Add ~29 new i18n keys to both EN and RU dictionaries:
  stars.notePrefix/Middle/Suffix, 12 month.* keys, 3 projects.group.*
  keys, 7 projects.desc.* keys, experience.item2Org
- Fix test: regex for fork-stars assertion tolerates span markup
- Fix test: filter Jinja template expressions from i18n key collection
- Bump i18n.js cache-busting version

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S8oQHBErmyHAWRedfFifZk

Drop the section-note disclaimer about 5 fork stars — merge all stars
into one total (108) for a cleaner presentation.

- Replace notePrefix/noteMiddle/noteSuffix with a single sectionNote key
- Use stats.stars_earned (108) instead of latest_total (103) in template
- Update currentLabel from 'own repositories' → 'all repositories'
- Update SVG desc to match the combined total
- Mirror changes in index.html, i18n.js (both EN and RU), and test

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S8oQHBErmyHAWRedfFifZk

…оектам'

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S8oQHBErmyHAWRedfFifZk

Closes two gaps found while verifying PR #42 against issue #41:

- The stars SVG chart's <title>/<desc> (the only screen-reader-accessible
  description of the chart, via aria-labelledby) stayed hardcoded in
  English regardless of the selected language.
- meta.description/meta.ogDescription baked PR/star/starred counts as
  dictionary literals, so switching language after a bot-sync run would
  overwrite the freshly-synced numbers in the HTML with stale ones from
  i18n.js.

Adds a small {placeholder} interpolation layer to applyTranslations via a
new data-i18n-vars attribute (JSON on the node), and a readProfileVars()
helper that sources meta description counts live from the existing
<span data-profile-value> nodes profile/sync/apply_site_fragments.py
already keeps current — so there's still one source of truth for those
numbers, not two.

- i18n.js: interpolate(), readVars(), readProfileVars(); stars.chartTitle/
  stars.chartDesc dictionary keys (en/ru); meta.description/ogDescription
  now use {prs}/{stars}/{starred} placeholders instead of literal numbers.
- index.html + profile/sync/templates/stars.html.j2: SVG <title>/<desc>
  wired with data-i18n-meta + data-i18n-vars, kept structurally identical
  per the bot-sync marker contract.
- tests/site-smoke.test.mjs: 3 new tests covering interpolation, the live
  profile-value substitution, and chart title/desc markup in both
  index.html and the template. 47/47 JS + 42/42 Python tests pass.
- Cache-busting version bumped for i18n.js.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S8oQHBErmyHAWRedfFifZk

Two high-severity findings from Codex adversarial review of PR #42:

1. If i18n.js never runs (blocked by an ad-blocker/CSP, network failure,
   or a stale/missing deploy artifact), the .lang-loading class added by
   the <head> bootstrap script for any non-English language was never
   removed — styles.css's `.lang-loading body { visibility: hidden; }`
   left the page permanently blank. Since RU is the default language,
   this meant a full blackout for most visitors on any such failure.
   Fixed with an independent setTimeout(2000ms) fail-open in the same
   bootstrap script, trading a possible flash of untranslated English
   for a page that's never permanently blank.

2. The stars section's copy ('all repositories', current-total swatch,
   and the new SVG <desc> from the previous commit) all state
   stats.stars_earned (original + fork stars), but the plotted polyline
   came from stars-history.json's daily entries, which — per its own
   documented scope ("Original axisrow repositories; forks excluded") —
   never include fork stars. The chart's endpoint (103) therefore never
   matched the number the page said it ended at (108). Since fork stars
   have no per-day history (GitHub's stargazer-events API doesn't expose
   one), chart_data() now takes a fork_stars offset applied uniformly to
   every plotted point, so the line's endpoint always equals
   stats_earned — the page's 'all repositories' framing becomes accurate
   rather than aspirational.

- index.html: fail-open setTimeout for the lang-loading guard.
- profile/sync/generate.py: chart_data(history, fork_stars=0); load_history
  threads cfg.stats.fork_stars through.
- profile/projects.json: _note updated to describe the offset instead of
  the old (now-fixed) discrepancy.
- profile/tests/test_generate.py: 2 new tests (offset applied to every
  point + ceiling scale, and default/explicit-zero parity with prior
  behavior); existing endpoint test extended to assert chart.latest_total
  == stats_earned.
- tests/site-smoke.test.mjs: new test pinning the fail-open setTimeout
  in the bootstrap script.
- 48/48 JS + 44/44 Python tests pass. Mutation-checked both fixes by
  temporarily reverting each and confirming the new tests catch the
  regression.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S8oQHBErmyHAWRedfFifZk