string.format must be a string according to
https://json-schema.org/understanding-json-schema/reference/string.html#format

We were using our own version of metaschema with a patch for this. Now
we are back to using official schema, but with a custom format
`internet-address` that should be equivalent to:

    "oneOf": [
      {
	"format": "hostname"
      },
      {
	"format": "ipv4"
      },
      {
	"format": "ipv6"
      }
    ],

The schema was not matching implementation (it was not even correct
schema)

Using
https://python-jsonschema.readthedocs.io/en/latest/validate/#versioned-validators
instead of a local copy of json schema

( not needed for the current work on jsonschema draft 7 support, but
done at the same time, to merge together two unrelated changes which
cause recompilation so that we recompile only once )

jsonschema = 3.0.0a3 needs a recent setuptools, probably because it uses
setuptools_scm

jsonschema needs six>=1.11.0

Version and requirements information containing six:
  [versions] constraint on six: 1.10.0
  Requirement of mock==2.0.0: six>=1.9
  Requirement of slapos.core==1.4.11: six
  Requirement of bcrypt==3.1.4: six>=1.4.1
  Requirement of cryptography==2.3.1: six>=1.4.1
  Requirement of jsonschema: six>=1.11.0
  Requirement of manuel: six
  Requirement of cliff: six>=1.9.0
  Requirement of pyOpenSSL>=18.0.0: six>=1.5.2

with support for draft7 of json schema

Problem has been found, and as this mode emits a lot of data with test result
disable it.

  The support to debian 6 was dropped, so this compatibility layer is not needed anymore.

  On OBS the build system is quite particular and it has to work in a specific way,
  This file handles the difference between build from source and prepare things to OBS.

It will allow to track behaviour of the test on various test nodes.

Drop not needed references and logic for ssl_ca_crt old implementation.

  This version only display informations to change keyboard and set a NDB.

In "caddy-frontend: Improve generated files" ssl_ca_crt was mistakenly used
for client authentication, which is in reality not supported nor needed.

It came from misunderstanding of apache configuration.

custom_domain and server-alias on given slave do not have to clash, and
can be deduplicated during request parameter analysis.

/reviewed-on !444

Since caddy-frontend: Support wildcards the wildcards are supported.

server-alias and custom_domain can be wildcards, so support such case.

/reviewed-on !446

The assertion was fixed incorrectly.

By setting maxDiff to None the test failures are more explanatory.

authorised --> authorized

Because of checking slave id in a whole string, slaves which shall not be
authorized has been put on authorized list.

Example: -frontend-authorized-slave-string == "custom_http",
         slave_id = "custom" has been authorized.

As test_https_url proves, the separation of url and https-url is correctly
implemented.

The case where test runner is not enabled got broken in
2a457867

This time, we had test to prevent regression.

/reviewed-on nexedi/slapos!440

So that it renders correctly on https://pypi.org/project/slapos.cookbook

I checked with :
 * `python setup.py checkdocs`
 * `python setup.py check --restructuredtext`

as described on https://stackoverflow.com/questions/16367770/my-rst-readme-is-not-formatted-on-pypi-python-org

/reviewed-on nexedi/slapos!441