Fix watchdog in the installer.

Fix watchdog in the installer.

We only had one watchdog running that triggered after 30 minutes. If
an unattended upgrade (e.g. started by sysupgrade(8)) took longer than
this in total, the machine would reboot half way through an upgrade.
The intention was that the watchdog would be reset after each set
download and after each set installation. But this never worked
correctly because the actual upgrade ran in a sub-shell and WDPID was
not visible.
To fix this we first need to export WDPID to make it visible in the
sub-shell. Then reset_watchdog was guarded by $UU && reset_watchdog,
but UU wasn't visible either. But we can't export it because we would
enter a loop. We can just use the fact that WDPID is not empty to
restart the watchdog.
Lastly the watchdog process would keep stderr and stdout open. This
made the tee(1) hang that is collecting the autoupgrade log that is
mailed to root.
As a simplification, we don't need to run the watchdog as a
co-process, we don't want to communicate with it, we can just run it
in the background.
Problem reported by stsp
With & OK deraadt
OK millert