Add a --system argument for system-wide installation

[PLPeeters]

This PR adds a --system argument for system-wide installation.

Since there doesn't seem to be a simple way to figure out the $(prefix) variable of the Git installation, I had to resort to a kind of hacky way to figure that one out in order to put the system-wide gitattributes file in the correct place (see the _get_system_gitconfig_folder() function).

I also fixed the path of the gitattributes file in the documentation when using the --global argument, which was incorrectly documented as being .git/info/attributes instead of ~/.config/git/attributes.

I also took the liberty of adding a few newlines to improve readability.

[PLPeeters]

I also just went ahead and added a fix and a test for the --status command which was apparently broken, along with some unification in the treatment of shell output (setting text=True everywhere and removing calls to .decode() that are no longer necessary with this). I can do a separate PR for that if you prefer, but since it intersects with my changes it was just easier for me to include it in here.

[kynan]

Thanks for your contribution! That's a useful feature.

Thanks also for the other fixes / cleanup and for adding a test.

Just a minor nit, otherwise I think this is good to go!

[kynan]

Thanks, I think this is good to go now!

[kynan]

Sorry, spoke too soon, this is not passing tests:

1. looks like the text parameter to check_output is 3.7+ only. Can you use universal_newlines instead?
2. we're getting lots of stray output: fatal: --local can only be used inside a git repository

[PLPeeters]

@kynan Changed the text parameter as requested.

The stray output was due to the git config call to fetch the extra keys. I changed it to a simple git config when nbstripout is called without the --system or --global parameters, since that seemed to make the most sense.

I also fixed the Git test since it didn't run properly on my system even though it looks fine on Travis. Apparently the version of Git on macOS seems to insist on having a newline at the end of the attributes file (it threw an *.txt is not a valid attribute name: .git/info/attributes:1 error on all the Git calls in that test for me).

[kynan]

Thanks, excellent work, really appreciate your contribution :)