The initial implementation of the events-executor contained a bug where the executor
would end up in an inconsistent state and stop processing interrupt/shutdown notifications.
Manually adding a node to the executor results in a) producing a notify waitable event
and b) refreshing the executor collections.
The inconsistent state would happen if the event was processed before the collections were
finished to be refreshed: the executor would pick up the event but be unable to process it.
This would leave the `notify_waitable_event_pushed_` flag to true, preventing additional
notify waitable events to be pushed.
The behavior is observable only under heavy load.

Signed-off-by: Alberto Soragna <alberto.soragna@gmail.com>

Signed-off-by: Alberto Soragna <alberto.soragna@gmail.com>

…ent optimizations

Signed-off-by: Alberto Soragna <alberto.soragna@gmail.com>

Signed-off-by: Alberto Soragna <alberto.soragna@gmail.com>

Signed-off-by: Alberto Soragna <alberto.soragna@gmail.com>

before they were in the SetUpTestCase and TearDownTestCase but we have tests that call rclcpp::shutdown, thus disrupting successive tests

Signed-off-by: Alberto Soragna <alberto.soragna@gmail.com>