Restore collecting all errors from all supervisors

[jhakala]

Martin tested causing errors that would appear in multiple supervisors and reported:

well no. LV1 listened to the 1st notification and decided to goToError.
Thus stops listening to the LV2's notifications.
I think we have been doing this since we use notification system

Today we had an issue that affected all of the HCAL partitions, but the error that appeared looked like the problem was localized to HO. We should restore the old behavior [before using the notification system to propagate supervisor errors, the SUPERVISOR_ERROR parameter and the error message sent to the level0 collected all the errors from all the supervisors.]

[kakwok]

The behaviour has changed since we stopped polling LV2's health from LV1.
#106
LV1 could probably still collect the error messages from LV2 when LV1 is in error.
I am not so sure for LV2 though.