[KNOWN ISSUE] Keep-alive implementation should be different

[leske42]

According to RFC 7230 6.5 and 9112 9.5

Servers will usually have some timeout value beyond which they will no longer maintain an inactive connection.

This means timeout should be reset after activity like a request from client, but now we have it absolute from the beginning.
Nginx apparently has 3 different setting for timeout: keepalive_requests for maximum number of requests processed before closing the connection (this is 1k by default), keepalive_time for maximum time one connection is sustained (this is 1h, for us currently 60 seconds), and keepalive_timeout for periods of inactivity (75 seconds).