[PATCHv3] nvme-mpath: delete disk after last connection

[Sagi Grimberg]

>> I really do not think we should change the mpath behaviors years after
>> first adding it.
>>
> 
> But only because no-one ever tested MD on nvme-multipath.
> It has been broken since day 1.

That is my assumption as well, and its a problem that by default
a pci device unplugs or a fabrics controller is lost and removed
the stacked mdraid on top stops functioning.

I still think we can fix that by changing the semantics of implicit
queue_if_no_path and restore it as an explicit opt-in.