[FEATURE] Implement reconnect logic and exponential backoff in Subscriber

[DmytroHavryshGoTo]

Summary

Make RedisStream::Subscriber.listen survive Redis restarts. Previously the listener crashed on any connection drop; now it logs the disconnect, retries with exponential backoff using PING to detect recovery, and resumes consumption. If Redis comes back without persistence (or the stream/group was wiped), NOGROUP is detected and the consumer groups are recreated automatically.

What changed

lib/redis_stream/subscriber.rb:

• Wrapped the xreadgroup loop with rescues for Redis::BaseConnectionError and Redis::CommandError.
• On connection error: call wait_for_reconnect, which probes PING with exponential backoff (0.5s → 30s cap, doubling) until Redis is reachable, then resumes the consume loop.
• On NOGROUP: call ensure_groups to recreate consumer groups on all streams, then resume.
• Other CommandErrors still propagate (e.g. WRONGTYPE).
• Group creation extracted to ensure_groups(streams, group), called both on startup and after NOGROUP.
• Diagnostic logging added at every state change (disconnect, each reconnect attempt + outcome, reconnect success with downtime, group creation, NOGROUP recovery). Logs go to stderr via warn with a [redis_stream] prefix.

PING is used (not xreadgroup) for recovery detection because xreadgroup blocks waiting for messages — without a probe that returns immediately, we couldn't distinguish "Redis is back but no traffic" from "Redis still down."

Behavior

Scenario	Before	After
Redis restart	Listener crashes	Logs disconnect, retries with backoff, resumes
Redis flushed / different instance (NOGROUP)	Listener crashes	Recreates groups, resumes
Network blip	Listener crashes	Reconnects (typically <1s)
WRONGTYPE or other command errors	Propagates	Propagates (unchanged)
Broken poison message handler	Propagates	Propagates (unchanged)

Backoff caps at 30s. Downtime is measured with Process::CLOCK_MONOTONIC (immune to NTP jumps) and reported in the reconnect log: reconnected to redis after 4 attempt(s); downtime 3.50s.

Sample output

[mesmerze]

could you please consider #12 🙏

[mesmerze]

INITIAL_RECONNECT_BACKOFF
MAX_RECONNECT_BACKOFF
🙏