Migrate Nokogiri::HTML to Nokogiri::HTML5

HTML5 parser matches browser parsing semantics for edge cases
(e.g. block elements inside <p> auto-close the paragraph per spec).
All inputs are well-formed kramdown output, so no behavioral change
on production content (verified via html_diff.py: zero diffs).

Changes:
- Switch all six Nokogiri::HTML / HTML.fragment calls to HTML5.
- Add test for HTML5 parse-tree shape when <figure> is nested in <p>.
- Update comment in FootnotePreviewInjector to match new parser name.

Author: agude

_plugins/src/infrastructure/footnote_preview_injector.rb

@@ -42,7 +42,7 @@ module Infrastructure
     # to strip the preview when reusing rendered HTML (feeds, plain-text, etc).
     #
     # The full HTML string is never round-tripped through Nokogiri to avoid
-    # DOCTYPE mangling: Nokogiri::HTML is used read-only for extraction;
+    # DOCTYPE mangling: Nokogiri::HTML5 is used read-only for extraction;
     # injection uses targeted string replacement on the original string.
     module FootnotePreviewInjector
       Text = Jekyll::Infrastructure::TextProcessingUtils
@@ -68,7 +68,7 @@ module FootnotePreviewInjector
       def self.inject(html)
         return html unless html.include?('class="footnote"')
 
-        doc = Nokogiri::HTML(html)
+        doc = Nokogiri::HTML5(html)
         footnote_map = build_footnote_map(doc)
         return html if footnote_map.empty?
 

_plugins/src/infrastructure/text_processing_utils.rb

@@ -22,7 +22,7 @@ module TextProcessingUtils
       def self.clean_text_from_html(html_content)
         return '' if html_content.nil? || html_content.strip.empty?
 
-        doc = Nokogiri::HTML(html_content.to_s)
+        doc = Nokogiri::HTML5(html_content.to_s)
         doc.xpath('//script | //style').remove # Remove script and style elements
         doc.text.gsub(/\s+/, ' ').strip # Get text from remaining, normalize whitespace
       end
@@ -118,7 +118,7 @@ def self.slugify(text)
       # hidden preview text never leaks into plain-text output.
       def self.strip_tags(html)
         cleaned = strip_link_previews(strip_footnote_previews(html))
-        Nokogiri::HTML.fragment(cleaned).text
+        Nokogiri::HTML5.fragment(cleaned).text
       end
 
       # Removes book-link hover-preview markup (see BookPreviewRenderer), including

_plugins/src/seo/seo_meta_generator.rb

@@ -120,7 +120,7 @@ def raw_title
       end
 
       def decode_html_entities(text)
-        Nokogiri::HTML.fragment(text.to_s).text
+        Nokogiri::HTML5.fragment(text.to_s).text
       end
 
       def site_title
@@ -168,7 +168,7 @@ def site_description
 
       def strip_html(text)
         cleaned = Infrastructure::TextProcessingUtils.strip_link_previews(text)
-        Nokogiri::HTML.fragment(cleaned).text.gsub(/\s+/, ' ').strip
+        Nokogiri::HTML5.fragment(cleaned).text.gsub(/\s+/, ' ').strip
       end
 
       # --- Image ---

_tests/src/infrastructure/test_footnote_preview_injector.rb

@@ -106,6 +106,23 @@ def test_inject_flattens_block_elements_to_classed_spans
     refute_match(/<(?:figure|blockquote|figcaption|p)\b/, preview)
   end
 
+  def test_inject_figure_nested_in_paragraph
+    # kramdown wraps cited-quote figures in a surrounding <p>. The HTML5
+    # parser auto-closes the <p> before <figure> (per spec), ejecting it
+    # as a sibling. flatten_blocks converts the resulting empty <p> tags
+    # to empty fnp-p spans (harmless — they render nothing). Verify the
+    # preview preserves all content regardless of the parse-tree shape.
+    body = '<p><figure class="cited-quote">' \
+           '<blockquote><p>The quote text.</p></blockquote>' \
+           '<figcaption>—<span class="citation">Author.</span></figcaption>' \
+           '</figure></p>'
+    result  = Injector.inject(footnote_html(id: 'note', footnote_body: body))
+    preview = extract_preview(result).to_s
+    refute_match(/<(?:figure|blockquote|figcaption|p)\b/, preview)
+    assert_includes preview, 'The quote text.'
+    assert_includes preview, 'Author.'
+  end
+
   def test_inject_citedquote_full_fidelity
     # The real citedquote structure: figure > blockquote + figcaption,
     # where figcaption holds span.citation with <cite> and <a>. Existing