Pdfrw

Fixes, enhancements, and new examples:

* Python 3.6 added to test matrix
* Proper unicode support for text strings in PDFs added
* buildxobj fixes allow better support creating form XObjects
out of compressed pages in some cases
* Compression fixes for Python 3+
* New subset_booklets.py example
* Bug with non-compressed indices into compressed object streams fixed
* Bug with distinguishing compressed object stream first objects fixed
* Better error reporting added for some invalid PDFs (e.g. when reading
past the end of file)
* Better scrubbing of old bookmark information when writing PDFs, to
remove dangling references
* Refactoring of pdfwriter, including updating API, to allow future
enhancements for things like incremental writing
* Minor tokenizer speedup
* Some flate decompressor bugs fixed
* Compression and decompression tests added
* Tests for new unicode handling added

Several fixes and new examples, including:

- Logging was improved (PR 44)
- PageMerge bug with multiple program runs was fixed (issue 50, PR 56 )
- PageMerge with state save/restore was fixed (issue 36)
- In-memory PDF support was improved (issues 38 and 59, PR 43)
- from pdfrw import * now works properly (issue 35)
- Added Python 3.5 to regression testing
- fancy_watermark.py example works with watermarks and pages
of differing sizes, and shows usage of the PageMerge class (issue 33)
- PdfString import slightly improved (but full Unicode support slated for release 0.4)

Several new features, including Python 3 support, as described in the README.