client: add failover broadcast listener and sentinel active probe

[chenshi5012]

Summary

This PR adds two complementary failover-detection enhancements to the Java client, addressing the client-side aspect of valkey-io/valkey#3881.

---

Changes

1. Cluster mode — FailoverListener in JedisClusterInfoCache

A new inner class FailoverListener subscribes to the +switch-master Pub/Sub channel on every start node. When the server-side patch (valkey-io/valkey#4340) publishes a failover notification, the listener immediately calls renewClusterSlots() to update the slot cache in real time.

Key design points:

• One daemon FailoverListener thread per start node.
• Exponential back-off on reconnect (1 s → 30 s cap).
• Performs an immediate slot refresh on (re-)connect to catch any failover that occurred while the listener was disconnected.
• Can be disabled via setFailoverListenerEnabled(false) before first use.
• Properly cleaned up in close().

Before this change: clients discover master role changes only through MOVED/ASK errors or periodic CLUSTER SLOTS refreshes, introducing latency proportional to the refresh interval.

After this change: clients react within milliseconds of the failover notification being published.

---

2. Sentinel mode — ActiveProbeTask in SentineledConnectionProvider

A single-threaded ScheduledExecutorService periodically calls SENTINEL GETMASTERADDRBYNAME on every sentinel. If the returned master address differs from the current pool target, initMaster() is called immediately.

Key design points:

• Default interval: 10 s (configurable via the new probePeriodMillis constructor parameter).
• Set probePeriodMillis <= 0 to disable the probe entirely.
• Iterates sentinels in order and stops at the first reachable one.
• Daemon thread, stopped in close().

Motivation: The existing SentinelListener relies on the Pub/Sub +switch-master event. If the subscription connection is disrupted (network partition, sentinel restart), the client may not receive the notification. The active probe provides a safety net that guarantees convergence within at most one probe interval.

---

Compatibility

• Fully backward-compatible. No existing constructor signatures are changed; new constructors are additive.
• The FailoverListener is started automatically after discoverClusterNodesAndSlots() and can be disabled.
• The active probe is opt-in via the new constructor; the default interval is 10 s.

Related

• Server-side PR: valkey-io/valkey#4340
• Issue: valkey-io/valkey#3881