fixed another leak in USB code

Ok, those of you coming with panics due to kmem exhaustion w/ USB, I
have fixed another leak.  For some reason I assumed that big blocks
were being deallocated upon free, not being put back on the freelist.
(Have I mentioned how much it sucks that the USB code it self has five
different allocators?)

As mentioned in the commit message, I did some testing, and a simple
bulk transfer over aue did not increase the devbuf memory usage, while
before this patch, I got it quickly over 20megs and growing.

Sorry for the breakage.

-- 
  John-Mark Gurney				Voice: +1 415 225 5579

     "All that I will do, has been done, All that I have, has not."