Public API for pdfplumber-clj: open PDFs and extract text, words, characters,
geometric objects, and tables as plain Clojure data.

Document loading and the error model. The PDFBox parse boundary lives here;
higher namespaces work with the returned PDDocument handle.

Bounding-box math and coordinate conversion.

The public coordinate system has a top-left origin (matching Python
`pdfplumber`): a bounding box is `[x0 top x1 bottom]` in PDF user-space points
with `x0 <= x1` and `top <= bottom`. PDFBox works in a bottom-left origin, so
conversion happens here and nowhere else. All extraction code must use these
helpers rather than open-coding the arithmetic.

Geometric object extraction (lines, rectangles, curves) via a
PDFGraphicsStreamEngine subclass.

PDFBox delivers path coordinates already transformed by the CTM into page
space (bottom-left origin); we collect painted subpaths and flip them to the
public top-left coordinate system. Only painted paths (stroked/filled) yield
objects; clip-only / no-paint paths are discarded.

Lightweight cropped page views. A view carries the document handle, a page
number, and a crop bbox; extraction functions accept it in place of a
document handle and restrict their output to the bbox. Nothing is copied or
translated — a view is just resolved into `:page`/`:bbox` options.

Table extraction. The `:lines` strategy reconstructs a grid from ruling lines
(explicit line objects plus rectangle edges): near-collinear edges are snapped
together, grid intersections are found, and cells are the rectangles whose
four corners are all intersections. Words are assigned to cells by center.

Character, word, and text extraction over PDFBox's PDFTextStripper.

PDFTextStripper already reports direction-adjusted coordinates in a top-left
origin, so char maps are built directly from `getXDirAdj`/`getYDirAdj` without
a page-height flip. Words are formed by clustering chars into lines (within
`:y-tolerance`) and splitting on horizontal gaps wider than `:x-tolerance`.