The firewalld direct interface is deprecated and will be removed in a
future release[1]. Recently IPv4 connectivity inside docker containers
on our runners broke and after some troubleshooting, the issue was
pinpointed to the start of the fail2ban service. We also had issues in
the past where sometimes firewalld had to be restarted after boot before
network connectivity worked in libvirt on our runners.

The issuse may be due to a bug in the way fail2ban use the direct
interface, a bug in firewalld or a combination thereof. Let's just avoid
the direct interface altogether and create a clean separation, with
firewalld handling the blocking and fail2ban maintaining the ipset.

[1] https://firewalld.org/documentation/man-pages/firewalld.direct.html

fail2ban role now protects postfix, dovecot and sshd. other roles can drop
configuration files into /etc/fail2ban/jail.d/*.local to enable fail2ban to
monitor it's service.

This bans all requests exceeding 1/min in a time period of 30 minutes.
This might be too harse and can be adjusted later.