pg-based distributed worker registry for reckon-gater

Provides cluster-wide worker registration and discovery using Erlang's built-in pg (process groups) module.

Cluster-Wide Discovery

Workers registered on any node are discoverable from all nodes:

- pg:join/3` broadcasts group membership to all connected nodes - `pg:get_members/2` returns PIDs from ALL nodes in the cluster - When a node fails, pg automatically removes its members from groups == Design Rationale == This implementation uses pg instead of Ra because: 1. **Simplicity** - pg is built into OTP, no external dependencies 2. **Sufficient consistency** - Gateway worker discovery doesnt require strong consistency; finding "any available worker" is fine 3. **No conflicts** - Avoids Ra cluster ID conflicts with Khepri stores 4. **Automatic cleanup** - pg handles node failures automatically

Eventual Consistency

pg provides eventual consistency. During network partitions, different nodes may briefly see different worker lists. This is acceptable for gateway workers since:

- Workers are stateless proxies to the event store - Requests are retried with exponential backoff - Any available worker can handle any request

Key Features

- Register gateway workers for specific stores - Cluster-wide worker discovery via pg groups - Automatic cleanup on worker death (local monitor) - Automatic cleanup on node failure (pg membership) - Load balancing via round-robin worker selection