Pyreadstat

* write user defined missing values, value labels for spss and stata
* write variable_display_width and variable_measure for spss
* breaking change: missing_user_values is now dict instead of set both for reading and writing.
* windows wheels build with appveyor + m2w64-toolchain. Wheels for python 2.7 not produced
anymore, instead setup.py raises an error if trying to compile with 2.7 on windows so that
pip will raise this error when trying to install from source distribution.
* column_names_to_labels: new member in metadata object, a dict of column names to labels
* setup.py: ordering source files to make the compilation reproducible

* added writing capability.

* MANIFEST.in changed so that windows dlls get into the source distribution, setup.py
changed so that those get installed only on windows but not unix.
* Implemented option disable_datetime_conversion for all reading functions in order
to skip date/datetime and time object conversion.

* Same as 0.2.4, just fixed issues in source distribution so that conda build
succeeds.

* encoding support for xpt files.
* support for user defined missing character values in sav.

* User defined values fixed for sas7bdat, implemented for sas7bcat and dta.
* Readstat sources updated.