Cancel upstream requests when clients go away

[ignoramous]

Sometimes, clients close connections even before a response could be sent. In those cases, it may be better to close upstream connections as there's no one to send an answer to, but then again if the client comes back again with the same query, we could possibly respond faster since the response would have been cached from previous request...

Eitherways, any behaviour, one way or the other, needs to be codified (documented or implemented).

[ignoramous]

Since we cache responses, it is okay for clients to go away... The caches would be primed for when they come back (assuming TTLs are within range).

Recent refactor has introduced timeouts too (e566bbd), that actually work. The min timeout is 4s while max is 30s (33e9107). 10s is the default timeout set on all platforms (1bdd319), though deployments are free to override those.