Add bounds to AsyncTransport, Ok, and Error

[jhillyerd]

Adding Send + Sync bounds to AsyncTransport and associated types (for #770)

This reduces the amount of bounds that must be specified for a function that is generic over AsyncTransport.

In my case:

pub async fn send_alert<T>(mailer: &T, subject: &str, body: &str) -> anyhow::Result<()>
where
    T: AsyncTransport + Send + Sync,
    <T as AsyncTransport>::Error: 'static + Send + Sync,
    <T as AsyncTransport>::Error: std::error::Error,
{

becomes

pub async fn send_alert<T>(mailer: &T, subject: &str, body: &str) -> anyhow::Result<()>
where
    T: AsyncTransport,
    <T as AsyncTransport>::Error: 'static,
    <T as AsyncTransport>::Error: std::error::Error,
{

I am not certain if adding the 'static and std:error bounds would be reasonable.

[paolobarbolini]

I'm all for adding 'static to Ok, 'static + StdError to Err. I don't think it'd be a problem to also have Send + Sync, but I'd first look into if there is a way of living without ever having to specify them anywhere

[jhillyerd]

Just to make sure I understand, you'd like to start with

adding 'static to Ok, 'static + StdError to Err

and revert the Send + Sync change for now?

[paolobarbolini]

Yes. I wouldn't necessarily revert Send + Sync though, but I'd look into if it's really necessary

[paolobarbolini]

I'm thinking of going forward with 0.10.0 without this as it's not a major issue and it will buy us more time to think about it

[jhillyerd]

I think that's fine, I'm not sure when I'll have time to really dig into this. I've learned a bit more about the expectations around Sync/Send, my current thinking is that forcing folks into Send may be a mistake if they need thread-local resources. Sync still seems like table stakes for an Async API though.