recoverying from servers that are down

[Chaskiel Grundman]

The server selection logic does not seem to actually attempt to
recover when a server is marked down.

I would think that afs_volume_pick_fileserver should have two passes
and use time_is_before_jiffies(server->fs_dead_jif) to decide to
ignore server->fs_state on the second pass and use the server anyway.
It appears that this might have been missed because if the server has
no refs, it will be reaped and the remebered state discarded.

Is there an afs development tree I should be using/basing patches off of?