engine: fix grace period

[edsiper]

In the current implementation when a shutdown has been requested through
a SIGTERM (or catched by ctrl-c), Fluent Bit initialized the shutdown
process and gives a grace period to wait for the active tasks to finish.

The problem with the approach is that grace period was not being respected,
actually the grace period was being renewed always waiting to flush all
the pending tasks.

This patch changes the behavior of the grace period where now the grace
period is considered (as it should) the 'maximum time to wait' for pending
tasks. If the tasks did not finished flushing under the flush period the
service will stop right away.

In addition, when no tasks exists, the patch makes the service stop before
the grace period time, since there is no need to continue waiting.

Common tests:

1. Exit before grace period

fluent-bit -i cpu -o stdout -f 1

hit ctrl-c, the service will stop before grace period (at second 1 or 2).

2. Force active Tasks by using an uresponsive remote network address

fluent-bit -i cpu -o stdout -m '*' -o http://192.168.3.4:3833 -m '*' -f 1

on SIGTERM, the tasks will still be active because the unresponsive network,
by forcing a shutdown the grace period will be respected and the service
stop at the right moment.

[l2dy]

Finally! Thank you.