If multiprocessing and concurrent unspported, render linear compilation.

Use generators and add a gzip support.

Signed-off-by: Timothée Peignier <timothee.peignier@tryphon.org>

Avoid listing all files for every finder for finding a file. This is
actually a bottleneck when the directory hierarchy is nested or the
number of files served in DEBUG mode is big. Since this flow is done in
every request, it slows down page loading.

In production environments where templates are usually cached performance
doesn't get hit by this.

Minor performance improvement, avoid memory relocation every time we
iterate and do a string concatenation. Instead add results to a list and
join them at the end.

References issue #282.

I added an import for `mark_safe` and wrapped the JavaScript / CSS
urls in it. All tests passed.

prevent pipeline from dying on Closure Compiler warnings

Coffeescript compiler now accepts .litcoffee and .coffee files.

fixed issue #271

Update using.rst

we started to use django-pipeline in Borsala production servers.

Fix Python 3 compatibility

Python 3 compatibility has been lost. Readd it.

finders.py: use chain to concatenate dictionaries instead of + operator.
glob.py: use list compresension instead of filter and lambda.
utils.py: use urllib.parse.quote instead urllib.quote if available.

Remove unneeded unittest2 dep.

Documenting the default JS template behavior w/ extra_context

Support for the async and defer tag attributes is not documented and requires examining the source.

Include documentation sources in PyPi releases.

It is not used anywhere.