On nodes with autoupdate disabled the cmdline parameter '-f' has no
effect. But using the autoupdater for manual updates is quite handy.

This will free about 5 to 12 MB of RAM even on lightly used devices
yielding plenty of RAM for autoupdate to succeed.

This makes gluon-autoupdater depend on coreutils-sha512sum when it's not
included in busybox.

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.

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 $(())

Makes manifest.sample use the model instead of the board name as that is
what the script actually expects. Also, a sample signature is added.

Recently the autoupdater contained a case statement to distinguish
between TP-Link WDR3600 and TP-Link WDR4300. This was left over from
previous development versions of autoupdater that relied on the board
name instead of the model. It is thus no longer needed and be removed
safely.

It is now possible to specific multiple branches like this:

config branch stable
  option url ...
  list pubkey ...
  ...

config branch testing
  ...

and select one in autoupdater.settings:

config autoupdater settings
  option branch "stable"