[loaders] Loads accept LTTextLines as top level pdfminer elements, which breaks things

[jstockwin]

In the loaders, we do

elements = [element for element in page if isinstance(element, LTTextContainer)]

however, I have an example PDF (which I can't share) where one of these elements is an LTTextLine.

LTTextBox and LTTextLine are both LTTextContainers.

We're expecting to get LTTextBoxes, containing LTTextLines, which in turn contain LTChars. In this case, we're only getting an LTTextLine containing an LTChar. This breaks the code, for example when trying to find the font, since we try to iterate through the 2nd level, which should be an LTTextLine, but is now an LTChar which is not iterable.

I think the fix here is to search for instances of LTTextBox instead of LTTextLine.