Pull Queries: Local host should immediately know lags and fail stale requests

[AlanConfluent]

Is your feature request related to a problem? Please describe.
If A is an active/standby for some table and B wants to forward requests to A, B currently relies on lag reporting to know how far behind A is. We currently allow B to issue requests to A during startup time, before A has reported its lag so as not to incur large error spikes during rolling bounces of hosts.

Unless we implement #4472 , so that B can do the enforcement in these situations, A should do maximum lag enforcement itself since it should always know its local lags.

Describe the solution you'd like
Host A, in the above description, should enforce maximum lags on local data and should fail pull queries if the data is too stale.