Eliminate raw HTML injection: convert ~kg-html/captions to native sx

Add shared/sx/html_to_sx.py (HTMLParser-based HTML→sx converter) and
update lexical_to_sx.py so HTML cards, markdown cards, and captions all
produce native sx expressions instead of opaque HTML strings.

- ~kg-html now wraps native sx children (editor can identify the block)
- New ~kg-md component for markdown card blocks
- Captions are sx expressions, not escaped HTML strings
- kg_cards.sx: replace (raw! caption) with direct caption rendering
- sx-editor.js: htmlToSx() via DOMParser, serializeInline for captions,
  _childrenSx for ~kg-html/~kg-md, new kg-md edit UI
- Migration script (blog/scripts/migrate_sx_html.py) to re-convert
  stored sx_content from lexical source

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

blog/scripts/migrate_sx_html.py

@@ -0,0 +1,89 @@
+#!/usr/bin/env python3
+"""
+Re-convert sx_content from lexical JSON to eliminate ~kg-html wrappers and
+raw caption strings.
+
+The updated lexical_to_sx converter now produces native sx expressions instead
+of (1) wrapping HTML/markdown cards in (~kg-html :html "...") and (2) storing
+captions as escaped HTML strings.  This script re-runs the conversion on all
+posts that already have sx_content, overwriting the old output.
+
+Usage:
+    cd blog && python3 scripts/migrate_sx_html.py [--dry-run]
+"""
+from __future__ import annotations
+
+import argparse
+import asyncio
+import sys
+
+from sqlalchemy import select, and_
+
+
+async def migrate(dry_run: bool = False) -> int:
+    from shared.db.session import get_session
+    from models.ghost_content import Post
+    from bp.blog.ghost.lexical_to_sx import lexical_to_sx
+
+    converted = 0
+    skipped = 0
+    errors = 0
+
+    async with get_session() as sess:
+        # All posts with lexical content (whether or not sx_content exists)
+        stmt = select(Post).where(
+            and_(
+                Post.lexical.isnot(None),
+                Post.lexical != "",
+            )
+        )
+        result = await sess.execute(stmt)
+        posts = result.scalars().all()
+
+        print(f"Found {len(posts)} posts with lexical content")
+
+        for post in posts:
+            try:
+                new_sx = lexical_to_sx(post.lexical)
+                if post.sx_content == new_sx:
+                    skipped += 1
+                    continue
+
+                if dry_run:
+                    old_has_kg = "~kg-html" in (post.sx_content or "")
+                    old_has_raw = "raw! caption" in (post.sx_content or "")
+                    markers = []
+                    if old_has_kg:
+                        markers.append("~kg-html")
+                    if old_has_raw:
+                        markers.append("raw-caption")
+                    tag = f" [{', '.join(markers)}]" if markers else ""
+                    print(f"  [DRY RUN] {post.slug}: {len(new_sx)} chars{tag}")
+                else:
+                    post.sx_content = new_sx
+                    print(f"  Converted: {post.slug} ({len(new_sx)} chars)")
+                converted += 1
+            except Exception as e:
+                print(f"  ERROR: {post.slug}: {e}", file=sys.stderr)
+                errors += 1
+
+        if not dry_run and converted:
+            await sess.commit()
+
+    print(f"\nDone: {converted} converted, {skipped} unchanged, {errors} errors")
+    return converted
+
+
+def main():
+    parser = argparse.ArgumentParser(
+        description="Re-convert sx_content to eliminate ~kg-html and raw captions"
+    )
+    parser.add_argument("--dry-run", action="store_true",
+                        help="Preview changes without writing to database")
+    args = parser.parse_args()
+
+    asyncio.run(migrate(dry_run=args.dry_run))
+
+
+if __name__ == "__main__":
+    main()