[ofw] use of io work items

[Sean Hefty]

changed subject...

>What's the difference between having 3000 requests queued up in your
>abstraction that keeps a work item queued per CPU until all are complete, or
>3000 work items queued to the thread pool directly? 

If the number of system threads > number of work items (which is likely true),
then other users of the system thread pool will be able to run simultaneous to
processing the 3000 requests.  (Although I changed the code to process 1 request
per system thread callback, which provides greater fairness, but is less
efficient.)  This assumes that the modify QP call blocks.

We need to come to a consensus on what approach to use before adding other
asynchronous handling to either the existing winverbs driver or a kernel
winverbs driver.

- Sean