Change default timeout to none

[yusefnapora]

Per conversation with @vyzo in slack, the global request timeouts were causing a lot of problems for operations that take awhile to finish (e.g. publish, etc).

We talked about removing the global timeout and making certain operations opt-in, for example fetching the remote id for a peer. I started to do that, but I couldn't come up with a compelling reason to keep the timeouts at all, versus just passing on the error from concat when its timeouts are exceeded.

With timeouts on the mcclient -> mcnode http connection, you get a confusing error message, and you may end up timing out before mcnode would have given you a more useful message. I'm now thinking that if we're going to have a timeout parameter, it should be sent as part of the HTTP request (e.g in a query param), so that concat can alter its timeout.

Is there something I'm missing here, where timeouts on the API requests buy us something useful?

[vyzo]

I think we should keep the option, and just default it to 0 (no timeout) instead of removing it altogether.

It can be useful in dealing with slow peers, for instance when doing a listPeers -i where we know if it takes longer than 5s it's almost certainly going to be an error in the client side and we prefer to just return promptly.

Also, we will want to use it to effect concat timeouts when we have this functionality.

[yusefnapora]

That's a very sensible plan; I like it 😄