Minor fixes for share fetch workflow

[Kaushik Raina (k-raina)]

Summary

• Move epoch-0 ack rejection before session_validate to prevent destroying existing sessions on malformed requests
• Add SHARE_SESSION_LIMIT_REACHED error when fetch session cache is full (default 2000 per KIP-932)
• Return INVALID_RECORD_STATE for acks on records whose lock expired, were re-acquired by another consumer, or are no longer in Acquired state
• Write per-partition AcknowledgeErrorCode in ShareFetch and ShareAcknowledge responses instead of hardcoded 0
• Validate ack batch ordering (ascending FirstOffsets, non-overlapping ranges) and reject with INVALID_REQUEST
• Handle per-offset AcknowledgeTypes arrays correctly instead of collapsing to last value
• Apply piggybacked acks before forgotten partition processing so acks on partitions being removed succeed
• Allow epoch=-1 (final fetch) to carry acks in Topics array; only reject ForgottenTopicsData
• Propagate broad errors to AcknowledgeErrorCode on all ack-bearing partitions per KIP-932 spec
• Add group.share.partition.max.record.locks (default 2000) to cap in-flight records per share-partition
• Add configurable share.auto.offset.reset with default "latest" per KIP-932
• Implement partition rotation across fetch requests to prevent starvation
• Archive in-flight records when log retention advances start_offset past SPSO
• Change UNKNOWN_TOPIC_OR_PARTITION from top-level error to per-partition error so valid partitions still get data

[confluent-cla-assistant]

🎉 All Contributor License Agreements have been signed. Ready to merge.
_{Please push an empty commit if you would like to re-run the checks to verify CLA status for all contributors.}