Unfortunately slave_title was put by mistake, it supposed to be slave_reference.

Use safe JSON serialisation/deserialisation, as otherwise unusual
slave_references can lead to issues and also character case is not kept.

Also care about case of log access user, which was undetected since
slave_reference in tests were always lowercase.

This reverts commit 1f91f19d.

Unfortunately due to way how profiles are mangled by jinja2, in some cases
the strings are becoming lowercased, so it just does not work.

It was not caught by tests, as no test has uppercase slave.

slave_title is dangerous, as it can contain any characters; it supposed
to be slave_reference.

Because there's no way anymore to pass the install prefix via an environment
variable, a new `make-install-extra` option (empty by default) is added so
that `make install` command line can be extended with `DESTDIR=...`.

/reviewed-on !524

Hardcoded partition names has been fixed after the shortening.

More changes are required to paths, as partition reference base has changed.

Also trafficserver starts in more cases, so move ERROR to OK state in some
promises.

At least trafficserver is known to not start on partitions on testnode like:

/srv/slapgrid/slappartX/srv/testnode/XXX/inst/test0-0/tmp/inst/TestSlaveGlobalDisableHttp2-1

so shorten the TestSlaveGlobalDisableHttp2 and others to simple T, which will result with path:

/srv/slapgrid/slappartX/srv/testnode/XXX/inst/test0-0/tmp/inst/T-1

Improved supervisord management is critical for maintenance of instances
between tests.

Since nexedi/slapos!392 we must format the json as done by `format-json`.

The tools for this are `format-json software/*/*.json` or `husky` ( nexedi/slapos!424 ).

/reviewed-on nexedi/slapos!522

fix json format

fix json format

Since Caddy 0.11.1, certificates has to match sites
(Thanks Luke)

trafficserver-cache-availability.py may pass randomly, as we do not have a
way during the test to really fill it.

As some of the nodes can lag behind, the system can be in state, that those
nodes will send inactive (also destroyed) slave publish information. Before
publishing it to master, check if each of slaves is really present on master.

Tasks:

 - [x] prove it really works on simulated environment
 - [x] check impact on massive simulated environment
 - [x] cover with a test (optionally)
 - [ ] check test results with this change

/reviewed-on nexedi/slapos!519

/cc @tomo @jm

/reviewed-on !520

OPENSSL_BINARY is not needed in the test case, only SQLBENCH_PATH has to be
provided, and perl shall be available in the PATH.

Notes

stack/erp5

 * The service-auto-approve-amount to default is set to 1, in order that the
   only needed service is automatically approved.  As caucase is accessed
   internally (on local IPv4) only partitions on the same server will access
   caucase.

/reviewed-on nexedi/slapos!518