Fix coordinate inversion when MediaBox origin is not (0, 0)

[apoorvdarshan]

Fixes #1332.

Root cause

_invert_box (in pdfplumber/page.py) flips PDF coordinates vertically so
the origin ends up in the top-left. It did so about the MediaBox height:

def _invert_box(box_raw, mb_height):
    x0, y0, x1, y1 = box_raw
    return (x0, mb_height - y1, x1, mb_height - y0)

with mb_height = mb_raw[3] - mb_raw[1]. Because it subtracts from the
height rather than from the raw top edge (mb_raw[3]), the MediaBox's
raw y-origin (mb_raw[1]) is never subtracted. For a MediaBox whose
lower-left is (0, 0) this is fine (mb_height == mb_raw[3]), but when
the MediaBox does not begin at y == 0 every top/bottom is shifted by
that origin, and page.mediabox / page.bbox start at a non-zero y.
(pdfminer.six already normalizes object coordinates to a (0, 0)-origin
page, so the flip must not re-introduce the origin.)

Reproduction

A PDF with a centered-origin MediaBox [-100 -200 100 200] containing a
single rectangle drawn at raw coordinates 10 20 30 40 re (raw span
x: 10..40, y: 20..60):

import pdfplumber
with pdfplumber.open("centered.pdf") as pdf:
    p = pdf.pages[0]
    print("mediabox:", p.mediabox)
    r = p.rects[0]
    print("rect top/bottom:", r["top"], r["bottom"])

Actual (before):

mediabox: (-100, 200, 100, 600)
rect top/bottom: 340.0 380.0

The page's top edge sits at top == 200 and mediabox[3] == 600 even
though the page is only 400 units tall — the whole page is shifted down
by the origin (200).

Corrected (after):

mediabox: (-100, 0, 100, 400)
rect top/bottom: 140.0 180.0

top now ranges over [0, height], as expected.

This also matches the report in #1332: raw MediaBox (-100, -200, 100, 200)
previously yielded (-100, 200, 100, 600).

The regression is also visible on the existing tests/pdfs/issue-1181.pdf
fixture, whose first page has MediaBox [0, 200, 420.9449, 585.2756]:

• Before: page.mediabox == (0, -200.0, 420.9449, 185.2756), and object
  top values were shifted -200 into negative territory
  (min(top) == -41.3).
• After: page.mediabox == (0, 0.0, 420.9449, 385.2756), and min(top)
  is 158.7 (all non-negative).

Fix

Flip about the raw upper-y edge (mb_raw[3]) instead of the height.
This subtracts the origin correctly and is a no-op for the common case
(where the MediaBox lower-left is (0, 0), so the top edge equals the
height). The change is confined to the two MediaBox/CropBox call sites in
Page.__init__; the x-offset behavior introduced for #1181 is preserved
(object x-coordinates and page.mediabox[0] still keep the raw
MediaBox x0), and downstream consumers (process_object,
point2coord) read the corrected origin from self.mediabox
automatically, so no other code needed to change.

Tests

Added tests/test_issues.py::test_issue_1332, which uses the existing
issue-1181.pdf fixture (its first page has a non-zero MediaBox
y-origin) and asserts that the inverted MediaBox begins at y == 0, that
no object has a negative top, and that a sample character's coordinates
are corrected. The test fails on the current code
(assert -200.0 == 0) and passes with the fix. The existing suite
continues to pass (python -m pytest, excluding the test_repair.py
cases that require a local Ghostscript install, which is unrelated to this
change). black, isort, and flake8 are clean on the changed files.

Disclosure: prepared with AI assistance; reviewed and verified locally.