Use failure instead of error_chain

[grigoryk]

This is an early WIP to get feedback.

So far the only non-mechanical bit was conversion of InvalidBinding, since it wraps a BindingError.

Eventually fixes #586.

[ncalexan]

This looks basically like I'd expect. I think we should do a little duplication (declare Result<T>, duplicate bail!) to minimize the initial changes, and then we can decide if we want to drop bail!.

I'll look at doing the db crate, which needs a lot of love to declare non-string errors anyway.

[ncalexan]

I'm not against declaring Result<T> = std::result::Result<T, errors::AlgebrizerError> to reduce the changes. Or even implementing bail! in each crate -- it's easy.

The Err(...)? pattern is odd to me -- I prefer the return Err(...) expression.

[rnewman]

Using ? adds an implicit into to do conversion of errors, which is a little more compact than return Err(some_err.into()) or return some_result.map_err(|e| e.into()).

[grigoryk]

This is bulk of the work. Needs another cleanup pass, and rebasing on top of master - query-parser is gone, algebrizer changed a bit, and #736 is about to land.

[grigoryk]

Tests pass! I'm not super happy with how parts of this turned out, but it's not a good idea to let this fall behind the other work. We can iterate on the particulars once/if we hit problems along the line.

[grigoryk]

@ncalexan do you mind giving this a lookover before I have to rebase it again? 😭