Can I vibecode PDF Expert?

Build me a local PDF reader and annotator to replace PDF Expert. Requirements:

- Python 3.12 + FastAPI serving one page of plain HTML and vanilla JS on http://127.0.0.1:5200.
  PDF_DIR in .env is the library: a sidebar lists its PDFs and opens one.
- Viewer: PDF.js (`pdfjs-dist`, vendored under /static) draws each page on a canvas with a text
  layer. Zoom, fit width, a page thumbnail strip, and a search box that highlights every hit.
- Markup tools: highlight, underline and strikethrough on selected text; sticky note; freehand ink
  with a colour and width picker; rectangle and arrow; free text box. The server writes each one
  with PyMuPDF as a real PDF annotation (`add_highlight_annot`, `add_ink_annot`, `add_text_annot`,
  `add_freetext_annot`), so Preview and Acrobat show the same marks.
- Saving: write to a temp file, keep one timestamped copy of the original in ./backups/, then
  replace the file. Undo removes the last annotation by its xref.
- Forms and signing: fill AcroForm fields on the page; draw a signature once, keep it as
  ./data/signature.png, and stamp it on any page at a chosen size.
- Pages: reorder, rotate, delete and extract pages from the thumbnail strip, and merge several PDFs
  picked in the sidebar.
- Scans: a button runs `ocrmypdf --skip-text` on the open file and reloads it; show a plain message
  when ocrmypdf isn't installed.
- Tests: `pytest` builds a sample PDF, adds every annotation type through the API, reopens it with
  PyMuPDF and asserts the count and subtype of each annotation.
- Single user, bound to 127.0.0.1. No accounts, no telemetry, local-first: nothing is uploaded.
- Out of scope: editing existing text in place, e-signature requests, Apple Pencil, cloud sync.
  README: install steps (`brew install ocrmypdf` or the apt equivalent) and a warning that the app
  writes into PDF_DIR, so point it at a copy first.

Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.