Support advanced format drives, i.e. ashift=xx

[GoogleCodeExporter]

Current large capacity hard disk tend to lie about their true physical sector 
size, claiming 512 byte sector while in fact using 4096 bytes.  This forces 
simple writes into expensive read-modify-write cycles, causing significant 
slowdown [1].

A technical clean solution (aside the drives stopping to lie) is difficult, as 
the blocksize should really be a per top-level vdev option, but that breaks the 
way properties are handled in ZFS.  While technically not fully correct, the 
approach taken by ZFSonLinux looks the least bad to me.  

I have started to port the ZFSonLinux approach to our much older code base.

[1] ZFSonLinux has performance data here: 
https://github.com/zfsonlinux/zfs/pull/280#issuecomment-1381910 

Original issue reported on code.google.com by [email protected] on 5 Aug 2012 at 8:19