Make Rocket an implementation detail of Nemesis

[lopopolo]

This PR builds upon what I learned in rwf2/Rocket#1019 and pulls the Rocket dependency inside nemesis crate.

Rocket is an implementation detail of using nemesis to serve a Rack app. This PR creates an interface on top of Rocket to abstract over the server implementation if I want to later switch to e.g. a Hyper backend.

The public interface to nemesis is a Builder struct that is used to configure routes for serving static assets and mounting Rack apps at a base path:

• Static assets are stored in Rocket managed state and served directly, using extension based ContentType detection.
• Rack apps are managed by two new constructs:
  • RackApp wraps a Value and Mrb where the Value is a Ruby object that responds to #call(env). RackApp implements ValueLike and exposes a typed interface around #call(env) that returns a nemesis Response.
  • Mount contains the information necessary to serve a RackApp at a prefix. It contains the base path mount point, a Fn that returns an interpreter and a Fn that returns a RackApp given an interpreter.

To handle a generic Rack application that responds to all paths and all HTTP methods, I introduce a RackHandler custom Rocket handler that generates all of the routes. The RackHandler contains a cloneable reference to the Mount.

Currently, nemesis only supports the shared nothing exec mode where there is a new single use interpreter created per request. In a followup PR I would like to add support for per-worker-per-app interpreters.

This PR significantly refactors the error handling in nemesis crate. I found I was repeating myself a lot by introducing new Error types in each module and wrapping them and adding From implementations. Instead, there is a global Error type in the nemesis crate root that implements all errors for all components in nemesis.

foolsgold crate is now fully implemented in terms of nemesis and has no direct dependency on Rocket.