Relax the error type in handle

[wiktor-k]

This allows clients to directly use ? with their error types instead of mapping everything to AgentError which obscures the error reason.

[wiktor-k]

@baloo If you'd like to take a look at this I'd much appreciate.

I've made this PR after working a bit with a downstream crate that uses it and the annoying fact that everything need to be mapped to AgentError and that variant erases what actually went wrong (nothing useful in logs).

I've also made it return Failure instead of abruptly closing.

[baloo]

That is a problem with the downstream crate that should have an error like:

enum Error {
   Agent(ssh_agent_lib::error::AgentError)
}

impl From<ssh_agent_lib::error::AgentError> for Error {
   fn from(e:ssh_agent_lib::error::AgentError) -> Error {
       Error::Agent(e)
   }
}

[wiktor-k]

That is a problem with the downstream crate that should have an error like:

enum Error {
   Agent(ssh_agent_lib::error::AgentError)
}

impl From<ssh_agent_lib::error::AgentError> for Error {
   fn from(e:ssh_agent_lib::error::AgentError) -> Error {
       Error::Agent(e)
   }
}

I think it's quite unlikely that they will ever want to convert from AgentError since if AgentError is raised in the handle function then it's bubbled up up to tokio::spawn and that's it - it's at the end of the call stack.

The only reason where the client can get a AgentError is via listen but this is also a function that doesn't really return if all goes well so it may be quite an edge case 🤔

[wiktor-k]

Wait, I thought this through. The Proto/IO variants can't happen here since this is `handle` function that returns from the client's code. That is: the Proto/IO can only happen if *the client* constructs that variant (which, I think, is outside of what you've been thinking of). Thinking about it for a bit longer the `User` variant was always a bit weird: it's from a client code but it doesn't have any details of what happened. I'm wondering if it should just be removed altogether. Or maybe you're thinking that the client code will do some parsing of SSH structures "on demand" and then return an AgentError::Proto? Phew :)

[baloo]

Oh yeah, you're right. But we can remove the AgentError::User altogether then right?

[wiktor-k]

Yup. Heh, this along with your RustCrypto PR will make the crate quite a bit leaner :)