Finish clustering infrastructure (libcluster + horde)

[bleuropa]

Context

Loomkin.Teams.Cluster, Teams.Distributed, and Teams.Migration were introduced in Epic 5.6 (541e4a4, 2026-02-28). The code is well-structured and tested (13 tests passing), but clustering is completely inert — the dependencies were never added and the setup docs were accidentally deleted.

See docs/clustering-status.md for the full audit.

Current state (~70% complete)

Working:

• Conditional startup via Cluster.enabled?() in Teams.Supervisor
• Graceful degradation — all Code.ensure_loaded? checks fall back to local-only
• Manager/migration route through Distributed API correctly
• Typed signals for node join/leave exist
• 13 comprehensive tests

Missing:

• Add {:libcluster, "~> 3.3"} and {:horde, "~> 0.9"} to mix.exs
• Add clustering config to config/dev.exs and config/runtime.exs
• Wire libcluster event hooks to Cluster.handle_node_join/1 / handle_node_leave/1
• Surface connected_nodes/0 and agent_count_per_node/0 somewhere (UI or monitoring)
• Remove or connect orphaned utility functions (local_node/0, etc.)
• Restore setup documentation (original docs/clustering-deps.md was deleted in 90b1f86)

Priority

Low — single-node is fine for current usage. Becomes important when scaling to multi-node production (Fly.io topology is already pre-coded).

Labels

infrastructure, tech-debt