Pip installation fixes

[david4096]

The last setup.py worked because of the presence of the generated files in the python directory, but didn't transfer over the _protocol_version properly. In changing the script to move over the file, I opted to use the python directory to include the generated python and just ignore the _pb2 files. This simplifies the installation by removing the need for a temporary directory, which is what was causing the problem. This also makes development easier. After checking out the tree, to generate the protobuf bindings, do python setup.py develop. This will add the generated files so you can run the tests.

This method of using a staging directory was also failing when using setup.py from another package.

You can install this PR and test it out using:

pip install git+git://github.com/david4096/schemas@pipfixes

I ran into this problem when working on ways of managing dependencies against git branches using setuptools and pip. The problem was that the setup.py call to install the schemas from server would execute the setup script for schemas, but the resulting package didn't contain the generated protobuf. When installing from pip this wasn't a problem. I understand pip to use setup.py to do its installation so I found this confusing.

In the end this PR keeps the development pattern within setup.py. Instead of packaging the compiled protobuf in a separate directory before packaging, you run setup.py develop (or install) to add it to the working python directory. This makes the contents of the resulting package more clear and development easier.

Compiling protocol buffers schema to python and putting it in a python namespace is a generally solvable problem. Perhaps we can package the process schemas script as a separate module.

[ejacox]

I tested it by adding git+git://github.com/david4096/schemas@pipfixes to requirements.txt from where I am working, e.g. server. +1