fix idle timeout issue

[rdallman]

details of bug are here: nodejs/node#13391
docs here: https://nodejs.org/dist/latest-v8.x/docs/api/http.html#http_server_keepalivetimeout

tl;dr is node set a 5s idle timeout. this does what go and java both seem to
do (have not checked python or ruby, though from what I know about python,
we are not doing this either) and doesn't have an idle timeout. since in this
case we do kinda trust that the client is using an idle timeout (it's fn),
this seems like the right policy anyway (if fn dies is something to consider,
but the least of our worries is fdk conns in that case, and we are killing fn
spawned containers on startup too). it should also be noted the client (fn) is
only using 1 conn per container.

[rdallman]

see fnproject/fn#1470 for the war story...

[denismakogon]

For Python FDK we have 75 seconds for keep-alive timeout

[rdallman]

@denismakogon can we bump up / turn it off?

[denismakogon]

I don't think we can turn off keep-alive timeout, because it doesn't work like in node.js, in python 0 means that connection must be closed right away, that's why i picked bigger number.

I'd say if someone will face the problem then we'd need to bump keep-alive timeout. As an alternative would be good to let (from Fn) FDK know which idle timeout out to use.

[rdallman]

i'm starting to think we should turn it off on the client side idle timeout and let the servers do it tbh... the client side we are closing them when we kill the container anyway, and the hole where we don't close them if fn dies is open whether we use idle timeout or not.

[rdallman]

actually, that won't fix either, since we might still use a conn on the client side that's been closed on the server side. fun.