Broker retry when server does not respond/connection break

When there are multiple replicas available, when one server failed to respond for a query, broker can retry with other servers (enabled by a flag to keep backward-compatible with the current behavior)

No files, tests, or entry points are named. Start by tracing broker query handling, server-failure detection, and replica selection; done means a flag-gated retry uses another available replica when a server does not respond or the connection breaks, while preserving current behavior when disabled.

Written by the indexing model from the issue text.