[Critical] Dashboard 5s list_sessions() poll decodes every session's full rmp blob to count rows

Severity: Critical
Domain: Performance

Location

crates/librefang-api/src/routes/config.rs:3031

substrate.list_sessions().map(|s| s.len())

Problem

list_sessions() returns Vec<Session> — each row's full rmp-encoded message history is decoded just to compute .len(). The dashboard polls this every 5 seconds (with refetchIntervalInBackground: false, so foreground-only — but every active operator hits it constantly).

At 100 sessions × 200 KB each, the daemon burns ~20 MB/s of needless decode work plus the rmp deserializer's allocation pressure, for what is morphologically a SELECT COUNT(*).

Fix

Use the existing count_sessions() (indexed SELECT COUNT(*)):

substrate.count_sessions()

Tests

• Bench (criterion or ad-hoc) showing the swap reduces CPU/alloc on the dashboard hot path by orders of magnitude with N≥50 sessions.
• Snapshot test asserting the route still returns the correct count.