Looping in shell is bad, as it needlessly starts several new PIDs each
iteration. Use a pipeline instead, with a trick.

Also, use unicode time for the events. We don't want to travel in time
whenever we travel in space.

Store approval requests into a file. Read and decide what approvals we
already got.

Why set sleep to background when we wait for its exit anyway?

When updater.sh received ctrl+c it exited but all its children stayed
alive. Shell corers given signals for subprocess. To solve this we must
send termination signal on our own when exit is detected.

Implemented new cli option --state-log as replacement of environment
variable UPDATER_ENABLE_STATE.

Previously, if we only downloaded packages but then discovered we can't
install them (because of collisions), an empty notification message has
been created. We avoid that now.

• Clean up state files from previous run, so errors and list of actions
  are not preserved.
• Give the updater longer time to run (in case of really slow
  connections and a lot of updates, for example).
• Enforce busybox's ps in case there's another one as well (to make sure
  the output doesn't change).

Allow uci-style bools in addition to true/false to disable (or not) the
updater. Minor, tested change, skipping review.

You can place files to internal directory in src/pkgupdate to embedded
them to finished binary file. Such files can be then accessed using uri
of format internal://

Move the check to the updater.sh script. This way it won't spam with
fake error saying that the updater is disabled (which is problematic,
but the user asked for it). Also, it is a bit less hacky and the user is
able to update manually when requested.

Limit the time for which certain commands (including the main pkgupdate)
are allowed to run.

Lock the state manipulation, correct it in case the updater terminates
incorrectly, etc.

To let nuci and others know about the current activity.

And some more support for running instead of the old updater.