Cluster primary node map is not refreshed on slot renewal, so primary-only commands go to demoted nodes after failover - #4692

JedisClusterInfoCacheTest, failover replayed through renewClusterSlots instead of discoverClusterNodesAndSlots:

Expected: map containing ["127.0.0.1:7001"->...]
     but: map was [<127.0.0.1:7000=...>]

And against the cluster-stable docker env after CLUSTER FAILOVER on :7483:

primaries after     : [localhost:7479, localhost:7480, localhost:7481]
FLUSHALL -> JedisBroadcastException ... (2 succeeded, 1 failed; first error: READONLY You can't write against a read only replica.)
   localhost:7479 -> JedisDataException: READONLY You can't write against a read only replica.

primaryNodesCache is only written by discoverClusterNodesAndSlots, which runs once at construction. Every later refresh (MOVED handling, the topology-refresh timer, refreshClusterTopology()) goes through discoverClusterSlots, which rebuilds slots and prunes dead nodes but leaves the primary map as it was, so ALL_SHARDS broadcasts, keyless writes and getConnection() keep targeting the pre-failover primaries. Rebuild the map in the same write-locked block, as the initial discovery already does. With the change the run above ends in [localhost:7480, localhost:7481, localhost:7483] and FLUSHALL -> OK.

Fixes #4691.