Question 1:
Raid is not needed for _distributed_ storage. You would still need traditional methods to preserve data if you are using LVM or NFS as a cinder backend.
Ceph is an example of a distributed storage backed for Cinder (block devices). You would not want to use raid with ceph.
Swifts object storage is another example of a distributed storage system where using raid actually decreases performance in some cases while no giving any benefit.
To sum up very briefly, if you are using a distributed storage backend, you don't use raid. If you are not, then you can use raid.
Question 2:
Not my area of expertise. Ill leave that to someone else.
Question 3-4:
This depends on how you configure things. It can be ephemeral existing only on the compute node while that instance is alive; it can also be persistent and be attached and detached from many instances.
It is all in how you configure the system.