Windows behavior fixed, with appveyor tests passing

[jeremycarroll]

This PR fixes issues on the Windows platform, particularly python2, revealed in the appveyor test failures.
In particular, encoding when making external calls with subprocess was usually incorrect.

Two specific changes on py2:

• use strings not unicode to add to the environment before making subprocess calls
• encode any non-ascii command line arguments (particularly →) in utf-8 before making the call.

A third change for both py2 and py3 is to change the file operations to text mode rather than binary mode, which simplifies handling of encoding.

In the tests, two further issues were addressed:

• incorrect handling when hg (or presumably git) is absent (this issue was not visible on appveyor, but locally to me)
• test_usage_string_fork changed to not explicit encode, and to use byte strings instead; xfailing in the most difficult case to fix: Windows on py3.

NB: I don't have a windows machine, so I am entirely reliant on appveyor for above analysis

[ekohl]

There is a universal_lines parameter which also fixes encoding issues. Perhaps that'd remove the need for xfail?

[c4urself]

I believe it's universal_newlines FWIW

[c4urself]

@jeremycarroll great work, thank you! let's try to see if we can remove the need for xfail -- if you haven't got the time then I'm OK with a TODO that we can work on later.

Also: if you could clean up the commits a bit that'd be pretty sweet as well.

Will merge on your indication.

[jeremycarroll]

OK - I will rebase onto a new branch, with cleaner commits, and try the newlines thing ... I don't think that was it: the error was occurring because the -h option prints out some text including that arrow →. Then this was being past from the subprocess back into the python process, where the confusion was happening.

[jeremycarroll]

This PR is replaced by a new one