The promise runner launched in a subprocess with the instance's python
redirects stderr to stdout in order to reserve stderr for propagating
error messages.

However `preexec_fn` is used to run `dropPrivileges` inside the child
process, so when `dropPrivileges` calls `logger.debug`, the logger
writes to stderr before the redirection is set up.

To fix this, the `preexec_fn` function is modified to apply the same
redirection first and revert it at the end. Reverting it avoids the
need to communicate the new file descriptor for the original stderr
stream to runpromise.py.

The vulnerability that this test was checking was removed from the
database, let's use another one that looks more likely to be removed.

See merge request nexedi/slapos.core!368

this option was only used in generate-signature-key of
slapos.libnetworkcache but it has been removed in
slapos.libnetworkcache@765d4737

   This is same API already present on compute node, the core diference here is that the Software Instance oddly store the value.

  This clean up code, and the instance should be be responsable for this conversion
  as it is all instance related parameters anyway.

   Help evaluate if the data is actually the last already

   This prevent us rely on SlapTool implementation to search the Software Installation for a given URL

  Create specific API for status suffixes, since this is a widely used by monitoring system

  Move to workflow tranistion the duty of reset/set last data, so it work regardless of slaptool.

   Introduce it as normal API of Software Instance since the structure is well-defined so we can move away from SlapTool

    This aims to define an kind of export feature used widely on SlapTool, which is eventually used by the node on the proper context. So it would be reusable accross multiples API's.

   Move away part of SlapTool code into a Mixin specific to fill and get Computer Information cache

both signature-private-key-file and signature_private_key_file works for
slapos.core but only signature-private-key-file works in slapcache and
slapos.libnetworkcache

This fixes a bug when the program output contain some ansi code, like
the \e[0;31m ansi code to display text in color on the terminal. These
characters are not allowed in XML so if they end up in the XML-RPC
messages this cause the underlying expat library to complain that the
stream is not well-formed (invalid token).

Fortunately, supervisor supports this and has a `strip_ansi` option that
we can set in the config to prevent these problems.

  It relies on Person_requestSupport now

We just relax the conditions to have an OS match.

 * consider raspbian as a debian
 * never set id

After the change to distro.linux_distribution() in
425f5bb7, raspbian is correctly detected
and id is set, see the execution of both functions on raspbian:

>>> platform.linux_distribution('', '', '', platform._supported_dists, 1)
('debian', '10.11', '')

>>> distro.linux_distribution(full_distribution_name=False)
('raspbian', '10', 'buster')

slapproxy, just like reference SlapOS Master, has to transmit important
partition parameters, in order to really support all features expected by the
software releases.

platform.linux_distribution() is deprecated since Python3.5 and it's
removed altogether in Python 3.8.

This more or less reverts 75b2d79c.

We don't need anymore the patch for Ubuntu as it is correctly supported
in distro module.

This commit reverts ddd77222.

The goal is to fix nexedi/slapos.buildout!25 (comment 148861)

os.environ is a special mapping object that transparently modifies the
actual environment and we should never make os.environ make point to
something else (see
nexedi/slapos.buildout!25 (comment 149333)
and https://docs.python.org/3/library/os.html#os.environ)

Instead change the environment only when calling SlapPopen. We will see
if the problems fixed by ddd77222 appear
again...