This commit add component needed for Optical Character recognition:
ocropy and its dependancies

The Ocropy egg on Pypi is empty, so this component have to be built from
source. The default neural network model is downloaded from the author
website and installed in python2.7/share/ocropus.

The patch is here for to different thing: one is adding an
OCROPY_MODEL_PATH to be able to specify the position of models, the
other is importing for the first time a module that compile C code from
python, allowing us to have the static library built before runtime and
preventing us from using gcc through zope.

More models are available here:
https://github.com/tmbdev/ocropy/wiki/Models

  If there is no error, the raise can be missleading.

  Allow http2 protocol be enable or disable by default
  Allow set certain configurations for MPM to control the number of process and threads.

Check that backup kvm disk is not corrupted

Pin erp5.util to version 0.4.49, with:

 * process timeout
 * xmlrpclib.Binary wrapping

Whitespaces matter to exporter.exclude:
- patterns must *not* start with a space
- file *must* end with a newline

Also, de-duplicate paths.

  In order to make an resilient backup, resilience stack wait for the internal backup finishes,
  before proceed to notify the pbs. This commit keeps the default behaviour (wait for 10 min w/o changes),
  however allows test software release adopt a much shorter waiting time (1min).

  With this commit we can accelerate resilience tests in several minutes.

  This fix the real location of the log after use logrotate stack.

Makes it easier to deploy a functional replication setup.

They closed, and downloads get redirected to a wiki page.

Stop breaking this tool for everyone developping on slapos.

  Apache verbose way too much with LogLevel info

It was not working with wrappers to work around shebang length limitation.

https://bitbucket.org/sirex/rubygemsrecipe/issues/9/gem-shebang-line-length-limitation

  Allow set software-root via parameters allow the webrunner inside a erp5testnode,
  recover the the previous builds from previous resilient tests, and only update (instead build from scratch)
  This is a typical usage on the webrunner (keep updating SR), so there is no
  problem on use this approach on erp5testnode tests.

Use the same exclude patterns (receiving end may want to exclude a file set
different from the time backup was exported, or may have a different set of
buildout-managed files which would be dangerous and misleading to
overwrite).

Do not apply exclusions to unrelated paths ("project", "proxy.db" and
"etc").
Simplify global exclusion file generation.
Add support for absolute and partition-relative exclusion paths
(previously, only partition-relative were supported).
Pipe exclusion file to rsync, to avoid an intermediate disk write.
Remove some directory prefix duplication by using PWD-relative paths
(except in one place, for runner-import.sh.jinja2 comparison).

So avoid spawning "cat" only for this.
Also, variable expansion quoting.

More similar changes will come later as part of larger changes.
Also, fail if
  cd "$backup_directory"
fails, as it is supposed to exist by that point.