Subgraph service: Don't instantiate a new pool for each request

[o0Ignition0o]

We're currently recreating a connector and a pool for each subgraph_service instantiation.

We shouldn't.
While we refactor it, let's make sure we don't forget to set a pool idle timeout duration.

[garypen]

We only build a new SubgraphService at startup and during re-configuration. This is done once for each subgraph, so each subgraph has its own http client (and associated pool). We don't build a SubgraphService for each request.

We may want to modify the code so that we only have one connector/pool to share across all subgraphs, but I think it's fine to have a separate connector/pool per subgraph, since we may want to vary behaviour between subgraphs at some point.

I'll add a pool idle timeout.

[bnjjj]

@garypen I think for every request we build a new subgraph_service, look here

[bnjjj]

Ok after some investigations we don't call SubgraphService::new on every request we're calling the SubgraphServiceFactory which clone the subgraph service. And cloning hyper::Client won't create a new pool, it's an Arc<Mutex<>> under the hood and so it doesn't create a new connection pool. We're good !