Add simulated disk Io

[mcches]

Is your feature request related to a problem? Please describe.
No. This is new functionality.

Describe the solution you'd like
Hosts have an Io concept today, but it is just network. Add the ability to write/read to/from disk, and have this state persist across host restarts.

[battesonb]

I'd like to pick this up. I have a partially implemented proof-of-concept on my fork from some tinkering this weekend using an in-memory virtual file system. I see a few "trophies" for getting this fully implemented.

1. Generic filesystem
   This is just a virtual filesystem over the generic parts of the API. The tokio tests can be grabbed and rewritten in turmoil style (example) -- although I'd suggest extra testing around the errors. There are some cases where this API bleeds into std and I don't really think there's a nice solution for that. For example the Metadata struct used by tokio is from std. In some cases we can use the std type, such as Path -- but it does have functions which won't work (like is_file).

2. File system interjections
   Similar to the hold and repair methods, we'd need some way to inject errors into the file system. I'm not sure whether this is needed for the first implementation, but separating them would require a version bump. I'd suggest separating simply because of the surface area of the fs module. A big-bang PR might not be the way to go for good feedback.

3. OS-specific implementations
   I currently have some of these just panic with the unimplemented!() macro, but a correct implementation would include these as well. For example, the different permission models across operating systems.

[mcches]

For (2), I don't think we need a version bump since this would be additive. It's probably totally different. We'll want an io failure rate with the same rng behavior as net and then the ability to corrupt files and stuff like that.