Possible fix for #207

in sys.path.  This ensures that lookups in __path__ will be the same
as sys.path resolution.

This also adds a replace argument to Distribution.insert_on meant
to be used with the replace argumen to WorkingSet.add.  This ensures
that new sys.path entries added via WorkingSet.add are inserted at
the beginning, rather than appended to the end.  This is necessary
for consistency with the above change, and kind of makes more sense
anyways.  This means that if a Distribution is added to a WorkingSet,
that replaces a different version of that Distribution, the new version
of that Distribution will have its location first on sys.path.

--HG--
branch : feature/issue-229

--HG--
branch : feature/issue-229

--HG--
branch : feature/issue-229

--HG--
branch : feature/issue-229

Based on experimentation, the canonical module name needs to be in sys.modules on Python prior to 3.3, but must be omitted on Python 3.3 and later.

--HG--
branch : feature/issue-229

--HG--
branch : feature/issue-229

Pop the module off the stack, preventing the 'Version' class from having a different manifestation in packaging than in pkg_resources.

--HG--
branch : feature/issue-229

--HG--
branch : feature/issue-229

Create a PEP 302 importer for managing conditional import of vendored packages from the 'extern' namespace. This technique avoids the use of 'imp' and works even when setuptools is installed as a zipped egg. Ref #229.

--HG--
branch : feature/issue-229

Move extern.packaging into a package to enable package-relative imports to resolve propertly. Ref #229.

--HG--
branch : feature/issue-229

Use the same technique in pkg_resources, relying on an 'extern' module to resolve the conditional import.

--HG--
branch : feature/issue-229

--HG--
branch : feature/issue-229

Modeling after Astropy's technique for bundling libraries, the imports are now much cleaner. Thanks @embray. Ref #229.

--HG--
branch : feature/issue-229

Fixes the original root cause of #231, and re-enables the test when the tempdir is a symlink (this does not explicitly test that /tmp itself is a symlink, but the effect is the same--only one of the path levels needs to be a symlink to reproduce this isssue)

Adds the regression test for distribute issue 323 that I attached to #207.  This is to ensure that any fix to #207 does not introduce another regression.

--HG--
branch : feature/issue-229

--HG--
branch : feature/issue-229