Pushed next wave of big changes

[Zach Brown <zab@zabbo.net>]

I just pushed a large series of commits that pivot the rpdfs architecture.  
There's three core elements that make up the pivot and unlock the rest
of the improvements.

First is having devd use a log-based storage format in the device.  This
requires mapping between block names and device location.  

Second, once we're mapping block names in the devices to begin with, we
can use large 192bit names for blocks on the wire instead of 64bit block
numbers.  This lets clients allocate new block names themselves without
having to negotiate access to global allocation structures. 

Third, we move from per-block cache states to per-inode locks.  This has
the benefit of communicating cache coherency with far less overhead but
at the cost of abdononing the (unneccesary, in the end) finer-grained
contention footprint.

The kernel client sees the most change from bringing these core features
online.  Large block names means we can reference blocks by their
logical identity rather than allocated mapped position.  We no longer
have block structurds that map logical to physical -- no btree for
metadata and no data mappings.  And with no global resource (block
numbers) all blocks are per-inode and we can easily use the vfs inode
and page caches for our caching.  

Anyway, it's all in the commits.  There's still a lot of work to do to
finish up the transition.  I wanted to err on the side of making this
work visible once the core pieces were there rather than waiting for it
to be more complete.

Enjoy!  More coming,

- z