Replicated read_log hangs when a cluster node is down (Studio times out) #172

A replicated read_log request hangs (and the Studio times out) when any node in the cluster is down. The expected behavior is to return logs from the reachable nodes and either omit or annotate the unreachable ones.

Reproduction

With one Fabric central manager stopped, the Studio's read_log call hangs:

{
  "operation": "read_log",
  "start": 0,
  "replicated": true,
  "limit": 100,
  "order": "desc"
}

Ask

• The replicated read_log fan-out should treat a node-down as a normal partial response, not a hang.
• Per-node response should include either the log entries or an error/timeout marker.
• A bounded per-node timeout so a slow/unreachable node can't stall the aggregate response.

Acceptance criteria

• With one node down, read_log replicated=true returns within a bounded time with logs from the reachable nodes and a clear indicator that the down node didn't respond.
• Studio renders the partial result rather than timing out the whole request.

Tracked in Jira: CORE-2960

🤖 Filed by Claude on behalf of Kris.