This is the initial version of our new configmode theme. It's far from
finished but at least configmode looks good. There are lots of
declarations copied from OpenWRT's default theme so most other luci
modules would be somewhat usable when shown using this theme.

The reset button resets the form. I don't think this is needed in a
wizard that simple and might confuse users so I removed it.

Due to my previous two commits configmode was broken when either meshvpn
or bandwidth limit was disabled. This fixes it but there may be a
cleaner solution provided by luci.

Hide the bandwidth limit checkbox when meshvpn is disabled.

Hide the downstream/upstream input fields when bandwitdh limiting is
disabled.

This way we have a kind of load balancing among the mirrors.

This adds a list of known models to the readme.

Instead of a single url this patch allows to have a list of urls. So in case
of network instability a router can still reach local mesh-cloud server to pull
a update from. In this stage the autoupdater simply tries every mirror until
one passes all tests.

This also removes the need for the cleanup function. Instead the cleanup
code is moved in place.

Move building blocks of the update into seperate functions.

gluon-autoupdater: get random number from /dev/urandom

The approach with awk's rand() wasn't really random between across
devices: When srand() was called without arguments, time() was used as
seed, which of course is the same on all devices when the script is
called via cron at HH:00:00.

This patch instead uses /dev/urandom as source of random (we don't need
cryptographically strong random numbers, so urandom is just fine) but
still uses awk for the comparison as busybox's ash cannot deal with
floats in $(())

The run frequency and exact time affect the alfred announce interval, so we can
just run it every minute to supply alfred with the most up-to-date data.

+ mgmt and forward counters
+ packet in addition to byte counters for everything

After a lenghty discussion, we settled on absolute vs. relative values.
Main reasons:

* stateless implementation on node possible
* convertable to relative values by differentiaion on receiver
* missed transmissions only decrease granularity, whereas relative
  values would introduce wrong numbers on integration if values are
  missed