making magic number-based detection of PDFs encoding-friendly, with tests

[jonoterc]

Magic number-based PDF-detection is vulnerable to encoding issues; forcing the first line to be read in binary mode should work across encodings (as the "%PDF" marker will be present using ASCII-7-compatible characters in any case). Also, removing the "end of line" anchor from the PDF version regex to avoid issues with non-printing characters in certain cases.

Added tests, details:

• extracted PDF detection from ensure_pdfs method for more direct testing
• new unit tests verify that PDFs without the ".pdf" extensions are detected as PDFs
  • samples are included under "test/fixtures/without_pdf_extension"
  • add further PDFs (without extensions) and they will be automatically included in tests
  • PDFs generated by Adobe InDesign previously raised encoding errors; samples (for each recent PDF version) are included
• new unit tests verify that files with the ".pdf" extensions are always detected as PDFs (regardless of file contents)
  • samples are included under "test/fixtures/with_pdf_extension"
  • samples file names reflect actual contents
  • not sure that this is the correct behavior, but not looking to change current approach

[jashkenas]

Nice!

[knowtheory]

Thanks for fixing this @jonoterc and sorry for the delay in merging it (and :heart: the additional tests). Just cut a release for it: https://rubygems.org/gems/docsplit/versions/0.7.5

[jonoterc]

Great, thanks!