Add ability to group collections of image bounding boxes

[edknv]

Description

Partially addresses #353. This PR is focused on PDFs. Other document types will be addressed in future PRs.

This PR adds support for a new extract_images_method option in the PDFium-based extraction workflow, allowing users to choose between a simple “as-is” extraction of image objects (simple) or a “grouping” approach (group) that groups related bounding boxes together. It introduces a new parameter, extract_images_params, for configuring thresholds and additional clustering details.

In the previous approach (simple), we only extracted PdfImageObjects, but PDFium provides several additional graphics objects in addition to PdfImageObject. For our purposes, we now also extract:

• The PdfPathObject FPDF_PAGEOBJ_PATH The PdfPathObject is an arbitrary shape made up of straight lines, rectangles, and cubic Bézier curves. A path may intersect itself and may have disconnected sections and holes.
  The PdfFormObject FPDF_PAGEOBJ_FORM is used to group graphical elements together as a unit for various purposes.

As an example, the following has a PdfFormObject. The object inside the red box is a form object that contains several nested component images. If we use the bounding box of the PdfFormObject (instead of naively extracting all the images with depth > 1), we can correctly group together all the components.

In the following example, we have what looks like an image, but it's actually made of many "shapes" (PdfPathObject). We can group them together as one single object.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[mara004]

Note that the docs linked above are for a commercial .NET SDK and not applicable to pypdfium2/pdfium.
The official docs are here:
https://pypdfium2.readthedocs.io/
https://pdfium.googlesource.com/pdfium/+/refs/heads/main/public/ (see the comments in the header files)

[mara004]

I'm not sure this method is generally correct, as the PDF spec allows for a pretty arbitrary coordinate system.
Indeed, the origin typically is the bottom left, but IIRC the coordinate system can be laid out between any opposite corners in principle. Also, the coordinate system does not necessarily have to start at 0, it could be any arbitrary value.

Related: pypdfium2-team/pypdfium2#284