As per @reinout comments in PR #253, let's make it clear that we're
assigning request._urlopener to a patched URL opener.

Behind a proxy, `urllib` does not work with Python2 due to [Python issue
24599](https://bugs.python.org/issue24599).

Instead of using `urllib`, let's implement `urlretrieve` with `urllib2`, which
instead works and leave the Python3 implementation untouched (since that
one works too).

This fixes buildout issue #32.

There's no need to change the `try: except:IOError` block where `urlretrieve` is
used (line 195 in download.py) since the exceptions raised by `urllib2`
are subclasses of `IOError` and will then be caught.

Better, simpler fix than #250

Bootstrap fixes, including develop-eggs removal

Buildout no longer breaks on packages with non-ascii filenames in them.

Slightly improved version of lelit's #172

This means the eggs/ cache dir creation also moved to the start of buildout
instead of happening somewhere deep in the installation mechanism. I'll need
to adjust the test output for this.

Avoid prefixing buildout dir to 'special' rpath elements.

Noted the change in buildout's changelog, too, as that's much more prominent
and as it fixes a buildout bug.

Replaces #204, fixes #30.

A date leads to less confusion than a version number separate
from buildout's own version number. Similarly, tracking buildout's version
number leads to bootstraps with a new version number but without changes or to
bootstraps with version number from an already-older buildout. So: a date is
best.

I don't think many people use this option as it isn't documented.
It used to be handy around the time when we had lots of breakage.
Nowadays it isn't needed.
So... I'm moving it out of the way to make room for --version as a proper version-returning option.

This is an urelated change from mgedmin's #204 pull request.
Prefixing is good as it might help figuring out what keeps filling up a temp
dir in case there's a problem.

See #86 and #222

Travis speed improvement through caching

The [:] is necessary to edit the os.walk results in-place. Otherwise the
filtering-out of .svn, CVS and so doesn't work as intended.
(Note: this regression I put in was caught by one of the tests! Good that
we've got such a large amount of tests!)

Added fs_to_text() function that ensures a filename is always unicode (python
2) or str (python 3).

os.walk() gets such a unicode dirname as start point, so that it tries to do
its best on python2. The resulting filenames are converted to unicode to make
really sure that the join() doesn't fail with a decoding error.

The input to the hash function is encoded, as it gives an ascii encoding error
otherwise on python 2.