Save HTTP requests to the registry and token manager logs in dedicated
log files and correctly logrotate them.

Also use a newer version of slapos.cookbook to properly revoke re6st
certificate when the slapos shared instance is destroyed.
Co-authored-by: Thomas Gambier <thomas.gambier@nexedi.com>

This partially reverts f834a06f.

This shebang was causing issues with testnodes. See
f834a06f (comment 228158)

This is a fixup of d2d0d49e.

We need to revoke the certificate in addition to remove the token.

See merge request !1724

- The 'wrapper' recipe is also extended to support them.
- The 'neoppod' recipe uses 'sig_ign' to avoid a potential kill by
  logrotate at startup.

See merge request !1724

See merge request !1724

Get check_gps_lock and check_frequency_outofbounds
promises from slapos.toolbox

See slapos.core!741

Apache failed to compile in environments in which the `which` command
was not present (such as Docker).
This is fixed upstream (by using `command -v` instead) in
https://github.com/apache/httpd/commit/05297d7b00f0cb1f20d3ff14f4409097df2615fe
but not yet released.

This commit patches the configure script in a similar way, so that
the `which` command is not used.

See merge request !1736

This allows to run multiple test suites on the same instances,
each test suite will wait until the current one is done

Latest fix for SSB NR ARFCN computation is needed for bands such as N79

qmpbackup 0.43 was leaving temporary files instead of removing them leading to
filling up disk space of the machines.

Cover the situation in test and pick up fixed version qmpbackup 0.44

Put the decolorization directly in the generated binary.

We don't create new instances anymore because it is currently
not possible to add instance nodes with slapos python client.
We therefore use existing services but set all necessary
parameters on them before each test.

TODO:
 * start and stop the services on each test
 * add a lock to prevent multiple E2E tests running on the
   same machine

If libiconv is not linked to gettext, I get the following error during
the compilation of gcc-10.5 (in a system with minimal requirements):

```
INFO msgfmt -o de.mo ../../.././libstdc++-v3/po/de.po
INFO msgfmt: Cannot convert from "ISO-8859-1" to "UTF-8". msgfmt relies on iconv(). This version was built without iconv().
```

See merge request !1732

and revert : component/curl: backport a fix for ".netrc parse error" with git

Add new measurements for bands B40, N77, N79

request was introduced by mistake in 9d8e5003

Texinfo 7 depends on perl.
If one does not have perl installed, the following error is raised:

```
INFO checking for perl... /usr/bin/perl
INFO checking Perl version and modules... no
INFO configure: error: perl >= 5.8.1 with Encode and Data::Dumper required by Texinfo.
```

See merge request !1728

If slave section depend on slave-log-directory-dict they are being reinstalled
on each change in this section, so instead directly provide a path.

slave-log-directory-dict is only used to pass directory information about all
slaves to slave introsection mechnism, write it down in the comment.

expert-backend-allow-downgrade-ssl allows to configure each node to downgrade
SSL negotiation with the backends to insecure, OpenSSL 1 style, ciphers and
algorithms.

It's implemented only per node as it shall be only used in some specific
circumstances and the cluster default (by default) is false. Such separation
allows to setup each node independently, as it requires node restart.

backend-haproxy is hashing backend-haproxy-wrapper:output in order to be
reactive on applied change to the configuration; this will result with node
restart after the configuration.

Note that the special test backend has been moved to different file in order
to start it with subprocess with altered environment; using multiprocessing
with spawn context didn't worked out, as it altered test process and thus
supervisord environment.

This is a fixup of 61c13f54.