fix(invoices): render invoice PDF with html-to-image for correct Persian

html2canvas re-implements text layout and mangled Persian RTL output (joined
words, broken spacing, reordered emails). Switch the rasterizer to html-to-image,
which renders through an SVG foreignObject using the browser's native text
engine, so Persian shaping/spacing/bidi and oklch() colors all come out right.
Peyda is embedded as base64 @font-face via fontEmbedCSS so html-to-image skips
its slow document-wide font scan. Drop the now-unused html2canvas dependency.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

frontend/package.json

@@ -16,7 +16,7 @@
     "axios": "^1.17.0",
     "clsx": "^2.1.0",
     "framer-motion": "^12.40.0",
-    "html2canvas": "^1.4.1",
+    "html-to-image": "^1.11.13",
     "jspdf": "^4.2.1",
     "lenis": "^1.3.23",
     "lucide-react": "^1.18.0",