Fix to_image() omitting filled AcroForm widget values (closes #1367)

[Cyberfilo]

Summary

`page.to_image()` (which goes through `pdfplumber.display.get_page_image`) currently opens a `pypdfium2.PdfDocument` and loads the page without initializing PDFium's form environment. Filled AcroForm field text is rendered via PDFium's form-rendering layer (`FPDF_FFLDraw`), which only runs when a form environment exists. Result: filled form-field text is missing from the PIL bitmap, even though every standard PDF viewer shows it.

This is purely a PDFium rasterization gap — pdfminer text extraction (`page.chars`, `extract_text()`) is unaffected.

Reproduction

import pdfplumber

with pdfplumber.open(\"filled_form.pdf\") as pdf:
    im = pdf.pages[0].to_image(resolution=150).original
    im.save(\"out.png\")  # filled field values absent

(Reproducer + sample PDF in #1367.)

Fix

One-line addition in `pdfplumber/display.py::get_page_image`:

pdfium_doc = pypdfium2.PdfDocument(src, password=password)
...
pdfium_doc.init_forms()  # <-- new
pdfium_page = pdfium_doc.get_page(page_ix)

Per pypdfium2's documentation, `init_forms()` must be called after open and before loading pages. On a document without a form it is a no-op, so this is safe for the common non-form case.

Test plan

No new test added because the bug is in the PDFium rendering path (binary output) and the existing test suite already covers `to_image()` for non-form PDFs — those continue to pass with the no-op `init_forms()` call. For form PDFs the existing tests cannot detect the regression visually; the rasterized bitmap differs only in a few hundred pixels per filled field.

Happy to add a pixel-comparison or perceptual-hash test against the sample PDF in #1367 if maintainers want — let me know.

CHANGELOG

Added under `## Unreleased`.

Closes #1367.