FPDF_LoadMemDocument() not working correctly?

[mara004]

While working on #57 / #58, I first tried to use FPDF_LoadMemDocument() for non-ascii filepaths on Windows.
For testing, I temporarily edited open_pdf() to read all files into memory and pass the data to FPDF_LoadMemDocument().
Then I ran the test suite, which showed issues with rendering: Several failed pixel assertions. From looking at the files in tests/output, I noticed that a lot of content was missing, especially texts.
This is weired, as one would expect all loading strategies to yield exactly the same output, and I think it did work correctly with previous PDFium versions, so upstream might have recently introduced a bug.

[mara004]

Closing as it's a PDFium issue.