Flate png filters (#114)
* Addition of Flate PNG reverse filters: Up, Average and Paeth
* Notes to run test-cases
* flate_png tests with filters
* Reverse filters are applied to every pixel except the first of each scanline.
Fix dict key.
* Apply filter to the whole scanline
* Reverting to original flate_png to assert all tests pass
* When flate_png_orig returns error, data is None
* Can't use print in python 3
* Can't use print in python 3
* import correct xrange based on python version
* During flate_png reconstruction, previous row must be the previous scanline reconstructed
* subfilter must read the same byte from pixel to its left: https://www.w3.org/TR/2003/REC-PNG-20031110/#9FtIntro
* Addition of tests for flate_png_impl: flate_png_impl doesn't do array to X conversion but instead it returns array.array('B')
* Fix flate_png_impl tests for filters 2 (sub), 3 (avg), and 4 (paeth) as they require at least 2 scanlines. Simplify UP filter.
* Use ord() when applicable
* Remove debugging code
* Tests for flate_png_impl using http://www.schaik.com/pngsuite/pngsuite_fil_png.html f01n2c08, f02n2c08, f03n2c08, f04n2c08
* Missing png.log files
* Fix file path for local .png.log files
* Simplify filters and add more tests
* Addition of Jupyter IPython notebook capable of rendering buffers (rasters) used/produced by test_flate_png.py tests: requires matplotlib
* Comments on ipython notebook
* Assert that flate_png_orig produced different output (yet correct looking PDF War of the Worlds) when png data was compressed and filtered using Sub (f=1). New flate_png also generates correct looking PDF War of the Worlds.
* Cleanup: remove flate_png_orig, update expected.txt checksums, and add basn0g08.png.log and its test
* Remove Encrypt check from test_roundtrip. If 'expected.txt' has a valid hash, the test is expected to pass and produce a file with same hash. Revert all PDFs with Encrypted content to 'skip' because the roundtrip PDF is 'blank'
