Remove failure

[tony-iqlusion]

Since the time failure was written, std::error::Error has gained many of the features it was originally useful for, most notably Error::source for type-erased error chains/cascades (with downcast
support to recover the concrete type, if desired), and on nightly, backtrace support.

This commit attempts to remove failure in a way that largely preserves the existing error handling, replacing failure::Context with an internal abscissa_core::error::Context type which provides similar functionality, and more or less replaces the previous abscissa_core::error::Error type.

Additionally, it defines a BoxError alias for type-erased errors based on ideas from this rust-internals thread:

https://internals.rust-lang.org/t/proposal-add-std-boxerror/10953/1

It's bounded as Send + Sync + 'static to help in multithreaded/async contexts.

I'm not convinced this is good as a steady state, but rather could also be a stepping stone towards moving to some more widely supported library, as discussed in #144.

[tony-iqlusion]

There's definitely a lot more possible work that could be done on the error types (e.g. I'd like to eventually refactor abscissa_core::error::framework => abscissa_core::framework and drop the leading Framework* on the error types, using framework::Error instead.

However, this change seems like it's going to be somewhat high impact across my existing apps, so I think it'd be make sense to do this, ship a release and get everything updated to be failure-free, then pursue further changes, in order to make the process more incremental and lessen the impact.

To that end, this commit doesn't introduce any new error handling libraries (as discussed in #144).

[tarcieri]

Here are the current auto trait impls for abscissa_core::error::Context based on the current state of this PR:

It's Send + Sync, which is good. I'm also wondering about UnwindSafe: would it be good to add a blanket impl for the case where the associated Kind is also UnwindSafe? That would also seem to depend on the unwind safety of BoxError, as well as Backtrace, which I'd have to investigate (so perhaps postpone that for another PR).

[tony-iqlusion]

The CLI previously relied on failure::Error, which was kind of a hack. This file is in-line with what's in the current boilerplate for an error type.