WIP: implement heartbeat to keep connection alive

[chaudum]

Summary of the changes / Why this is an improvement

@autophagy @mfussenegger Is the keep-alive feature of crash still a thing? It has been lying around on my local checkout for some time now and I am wondering whether I should get it over the finish line.

Checklist

• CLA is signed

[mfussenegger]

I'm not sure. psql suffers from the same problem when using it to connect to a cloud cluster. So I wonder if we can come up a solution that doesn't involve patching all clients.

[chaudum]

I'm not sure. psql suffers from the same problem when using it to connect to a cloud cluster. So I wonder if we can come up a solution that doesn't involve patching all clients.

The very specific problem, however, occurs when using Azure Loadbalancer, which has an idle timeout of 4-30 minutes, but it is not possible to deactivate connection reset.
https://docs.microsoft.com/en-us/azure/load-balancer/load-balancer-tcp-idle-timeout#tcp-idle-timeout
load-balancer/load-balancer-tcp-reset

[mfussenegger]

Instead of doing query-heartbeats we could also enable TCP-keep-alive.

urllib3 docs has an example how to set that

[chaudum]

Instead of doing query-heartbeats we could also enable TCP-keep-alive.

urllib3 docs has an example how to set that

I will open an PR on crate-python.