[yarn] Correct username is not used on hard failover

When the master RM crashes, we have poped the RM api and we create a new one and push this later back into the pool?

I think there might be a flaw in the

def rm_ha(funct):

when we get the RM api from:

def get_next_ha_yarncluster():

this one does:
rm.setuser(DEFAULT_USER)

so we should probably reset it to the current user in def rm_ha(funct) ?