Bounding boxes are incorrect for rotated layout items

[dhdaines]

pdfminer.six reports the boundaries of glyphs in LTChar like this:

(x0, y0) = apply_matrix_pt(self.matrix, bbox_lower_left)                                                                                                                                                   
(x1, y1) = apply_matrix_pt(self.matrix, bbox_upper_right) 
# ... normalize them
LTComponent.__init__(self, (x0, y0, x1, y1))                                                                                                                                         

This gives the two opposite corners of the glyph, but these do not define the bounding box when the text matrix (or the CTM) contains rotation, in which case it's necessary to use all four corners. To illustrate, take this PDF (please) and visualize what pdfminer.six gives (yes, the code is very long! it's at the bottom...):

It should probably look like this:

This is also true for images (and maybe other things). You can fairly easily detect whether the bounding box needs to be calculated with all four corners because either b or c in the transformation matrix are negative (this means that when Y increases, X decreases, or vice versa, and thus the upper left / bottom right corners will be further left / further down), for example: https://github.com/dhdaines/playa/blob/main/playa/utils.py#L272

Code to generate samples above:

from pdfminer.converter import PDFPageAggregator                                                                                                                                                           
from pdfminer.pdfdocument import PDFDocument                                                                                                                                                               
from pdfminer.pdfinterp import PDFPageInterpreter, PDFResourceManager                                                                                                                                      
from pdfminer.pdfpage import PDFPage                                                                                                                                                                       
from pdfminer.pdfparser import PDFParser
from pdfminer.layout import LTChar
import pdfplumber
from pdfplumber.utils import bbox_to_rect
rsrc = PDFResourceManager()                                                                                                                                                                                
agg = PDFPageAggregator(rsrc, pageno=1)                                                                                                                                                                    
interp = PDFPageInterpreter(rsrc, agg)                                                                                                                                                                     
pdf = PDFDocument(PDFParser(open("samples/rotated.pdf", "rb")))                                                                                                                                                      
pdfpage = next(PDFPage.create_pages(pdf))                                                                                                                                                
interp.process_page(pdfpage)
height = pdfpage.mediabox[3]
layout = agg.result
glyphs = []
for item in layout:
    if isinstance(item, LTChar):
        x0, x1, y0, y1 = item.x0, item.x1, height - item.y0, height - item.y1
        if y1 < y0:
            y0, y1 = y1, y0
        glyphs.append(bbox_to_rect((x0, y0, x1, y1)))
pp = pdfplumber.open("samples/rotated.pdf")
img = pp.pages[0].to_image()
img.draw_rects(glyphs)