doc/admin: clarify that upgrades require 10m of downtime by default

Created by: slimsag

Prior to this commit we advertised that HA cluster deployment upgrades were zero-downtime. In practice, this is not true as database migrations may mean that e.g. some frontend replicas during a rolling updates will be unable to write (and sometimes read) from the database. In practice this is most likely a service-degradation situation, not complete downtime, but it is good to be explicit about planning a 10m downtime window.

For Sourcegraph.com, we will have zero-downtime upgrades with the explicit tradeoff of more rapid deployments. I am not documenting that here and am merely documenting the current state of the world. @efritz owns this.

This change was discussed and approved live with @efritz @tsenart @pecigonzalo @beyang