Tesserocr

* Support for Tesseract 4
* New OCR engines `LSTM_ONLY` and `TESSERACT_LSTM_COMBINED`
* New default tessdata path handling (104)
* Fixed compilation against Tesseract v3.05.02 which required c++11 (120)
* Fallback to 'eng' as default language when default language returned by the API is empty (103)
* Added notes about [Windows installation](https://github.com/sirfz/tesserocr#windows) in the README (97)

* Support timeout in Recognize API methods (55)
* You can now pass a `timeout` parameter (milliseconds) to the `Recognize` and `RecognizeForChopTest` API methods.
* Fixed typo in `_Enum` initialization error message formatting (56)
* Display tessdata path in init exception message (60)
* Fixed version check in Python 3 when reading the version number from the tesseract executable (60)

Fixed setup bug that affects gcc versions with no `-std=c++11` option support (which should be required by tesseract 4.0+ and not older versions). 53

* Improved setup script
* Tesseract 4.0 support:
* Two new `OEM` enums: `OEM.LSTM_ONLY` and `OEM.TESSERACT_LSTM_COMBINED` (tesseract 4.0+)
* Two new API methods: `GetTSVText` and `DetectOrientationScript` (tesseract 4.0+)
* `PyTessBaseApi.__init__` now accepts a new attribute `oem` (OCR engine mode: `OEM.DEFAULT` by default).
* `file_to_text` and `image_to_text` functions now also accept the `oem` attribute as above.
* Fixed segfault on API `Init*` failure
* Fixed segfault when `pixa_to_list` returns `NULL`
* Documentation fixes and other minor improvments

Bug fix release:
- Improved setup: attempt compile with default environment variables even if `pkg-config` fails
- `WordFontAttributes` now returns `None` instead of segfaulting when `NULL` pointer is returned by API

- Avoid crash for unexpected cflags returned by pkg-config (14)
- Added note in README regarding multiple tesseract installations (9)