Cater to IO latency sensitive volumes

[hickeng]

Breaking these aspects out from #159. Some workloads, databases in particular, require IO and latency guarantees and using sparse disks or delta disks may impact latency in hard to predict ways depending on low level block layout of the entire disk chain.

We should allow for the following when creating volumes:

• [ ] inflated size - let a user specify a capacity that is pre-inflated, trading off datastore space utilization for consistency.
• [ ] base volume - rather than share a base filesystem layer for creation performance, format the volume directly to eliminate delta disk overheads. This may be irrelevant as I fully expect the fast volume create to be used primarily for anonymous volumes created to back VOLUME commands in dockerfiles. I do not expect to need the same fast path for volumes created explicitly via docker volume create.
• [ ] raw volumes - pass through mechanism. I suspect this is something an admin would have to create and assign to the VCH via docker-machine, but here for completeness and to prompt thought when we come to this.

[mdubya66]

@fdawg4l still relevant?

[fdawg4l]

@hickeng asked for these features and I'm not sure customers or PMs want them at this stage of development. If it's my call, I'd move this to icebox (or whatever it's called now) and we can refer to the items when and if they come up or close it and reopen the items as individual stories (again as they come up in the roadmap).