[kv] Use HyperClockCache for the shared RocksDB block cache

[platinumhamburg]

Purpose

Closes #4610.

When shared block caching is enabled, all KV tablets in a TabletServer currently use one LRU cache. Use RocksDB's HyperClockCache to reduce cache-metadata contention under concurrent access.

Brief change log

• Create the shared block cache with automatic entry-charge estimation, default sharding, and a non-strict capacity limit.
• Exercise HCC in the existing two-database shared-cache test, including reads after one database closes.
• Preserve existing capacity configuration, metrics, and ownership. Independent block caches and WriteBufferManager accounting remain LRU; no new configuration option is introduced.

Tests

• ./mvnw verify -pl fluss-server -am -Dtest=KvManagerTest,RocksDBResourceContainerTest -Dsurefire.failIfNoSpecifiedTests=false -DfailIfNoTests=false (Java 11).
• Workload-level performance comparisons have not been run; no throughput or latency improvement is claimed.

API and Format

No public API or storage-format changes.

Documentation

No new configuration or user-facing API.