Fix OCR PDF converter: preserve inter-word spacing on image pages - #2666

Fixes microsoft#2565.

When a PDF page contained embedded images, the OCR path rebuilt each
text line by iterating page.chars (sorted by y then x) and joining with
an empty string. That glued adjacent columns together (e.g.
'Customer Name' and 'Vendor Name Ltd' became 'Customer NameVendor Name
Ltd') because chars from neighboring columns that happened to share a y
coordinate were concatenated without any space between them.

Use pdfplumber.page.Page.extract_text_lines(strip=False) for line
reconstruction instead, which applies pdfplumber's own word/column
spacing logic (the same logic used by page.extract_text()). Keep the
existing extract_text() fallback if extract_text_lines is unavailable
or returns nothing. Update the complex-layout test fixture expectation:
it was asserting the buggy glued form ('ItemQuantity', 'Widget A5')
and now expects the correctly spaced output ('Item Quantity',
'Widget A 5').