Running test as last makes no sense nowadays, as the site is fully prepared.
Awaiting for each possible log of configured frontend is not worth the
requirement, so make those tests as first ones, which simplifies running only
them with reproducible effect.

All tests which could be affected by that fact has been identified and changed.

otherwise it probably uses system's openssl.

Setting rpath seems not needed because we are using
-DBUILD_SHARED_LIBS=OFF here.

https://github.com/sigoden/dufs

New software to serve static files

See merge request nexedi/slapos!1270

caddy-frontend-N is bad name, it's just frontend-node-N.

This option was useful only during the time, when there were experiments
running on caddy-frontend clusters during switch to HTTP/2.

Currently HTTP/2 is a standard, and there is no reason to disable it globally.

Attention, this change will require manual intervention on the upgraded
cluster with:

sed -i 's/^recipe = caddyprofiledeps/recipe = software/g' .installed*

On each partition.

NAME_BASE is set on top of the file with informative warning, so that
developers are aware of consequences.

instance-slave-caddy-simplified-input-schema.json has been removed, as it is
not useful.

Some profiles were templates and vice versa, which was very
misleading.

The most complex case - no automatic https redirection and usage
of url and https-url is used to check implementation of type:redirect now.

haufe.requestmonitoring already supports WSGI and Python3 and seems
used and maintainted.