Generalizes the header write timeout to a broader "safety valve" write timeout.

[slackpad]

This opens up the timeout to the user's payloads as well as just control headers. It's a little indirect because we don't have a timeout in the send() goroutine directly, but this timeout effectively covers that because the caller is waiting for that goroutine to put the write error onto the error feedback channel.

[armon]

@slackpad Should we instead try to cast the connection to something with a SetWriteDeadline and use deadlines instead? That way we can save a ton of timer setup/teardown throughout.

[slackpad]

@armon We could do that by moving the timeouts into session.go's send() loop, but then we don't cover the cases where the sendCh itself is filled up. I think we want timeout protection on those as well.

[armon]

Ah makes sense! 👍