Pybufr-ecmwf

-fix a few python3 compatibility issues
-fix issue 4 (warning on max bufr message size had wrong threshold)
-add a simple example program to request a name for a descriptor number
-make bufr table copying code more efficient

-move to ECMWF BUFR version bufrdc_000400
-reorganise custom fortran code added to the ECMWF library
to allow easier interfacing by this module
-split test data from example_programs dir, and move to separate test dir
-fix a number of python3 compatibility issues

Contains mainly small bug fixes and updates the build script to the
latest library version.

Several fixes have been added in the handling of the BUFR tables, which
now should also work if you provide custom BUFR table names or a custom
BUFR tables directory (as parameter or through the BUFR_TABLES environment
setting) to the module.
In addition functionality to create new BUFR tables from scratch has been
added, including an example script to demonstrate this.

The top level interface for reading a BUFR message is in place now.
Still the code only works properly when you use the BUFR tables
provided by the ECMWF BUFR library.
Compilation with python3 seems to have some difficulties.
The dependency on numpy is no problem anymore, since that one is now
fully ported to python3, but the f2py tool that comes with numpy
seems broken somehow, and crashes during the build process.
The problem has been reported to the numpy developers (see
http://projects.scipy.org/numpy/ticket/1932), so for the moment
I'll wait for a bugfix from the numpy team before continuing
the python3 porting effort.

A next wrapper interface is in place now, and a start for the top-level
interface is implemented.
Also, the module can now be converted and compiled using python3.