For now just sets the 'float_is_vm' fact.

Retry dataset restores on failure

See merge request !269

On the first run the backup s3 bucket isn't set up correctly yet, so
we must restart the litestream-enabled services at the end of the
playbook (and tolerate their earlier failure).

Using the variable sso_auth_session_lifetime (ttl in seconds).

Parameterize local CA params in util-credentials role

See merge request !268

Potentially allows the usage of multiple CAs (for different contexts),
to be used extending the service_credentials attribute (not
implemented yet).

The rsyslog daemon will run inside a container, so the mount can be
parameterized instead.

The litestream restore script was empty even when the
proper configuration was defined, due to a bad check.

Consider empty public_endpoint.domains list to be the same as unset

See merge request !266

Renaming all float-related prometheus targets meant that the
"rsyslog.*" job filter was no longer working.

Add a NGINX config option to set X-Robots-Tag:noindex on specific URLs

See merge request !265

Another case of 'if is bad' in NGINX config. Better to just use
a map with a default empty value.

Support new syntax in haproxy

See merge request !258

Classic shell scripting
mistake (http://mywiki.wooledge.org/BashFAQ/050 for instance).

Fix breakage when docker_registry_url is unset.