fix(inkless:consolidation): DeleteRecords across tiers on consolidating topics

[viktorsomogyi]

Makes DeleteRecords correct end-to-end for switched consolidating diskless topics (both diskless.enable and remote.storage.enable), where the client-facing leader is a hash/AZ-selected replica (via the metadata transformer) rather than the real KRaft leader, and a freshly-rebuilt leader's local log start is pinned at the classic-to-diskless seal.

Changes, in layers:

• Routing (local leg → real leader). DisklessDeleteRecordsForwarder forwards the local leg of a DeleteRecords request to the actual leader, so the admin request no longer loops on NOT_LEADER_OR_FOLLOWER against a transformer-advertised follower. Wired through KafkaApis/KafkaApisBuilder/BrokerServer. DeleteRecordsCommand now exits non-zero on per-partition errors.
• Cross-tier low watermark. For consolidating partitions the completion low watermark and ListOffsets(EARLIEST) come from the authoritative control-plane cross-tier earliest, not the ISR (whose follower log starts are frozen at the switch). DisklessFetchOffsetRouter routes EARLIEST to the control plane (Problem A: cross-broker earliest consistency), and DisklessLeaderEndPoint reports min(crossTierEarliest, localLogStart) as the whole-log start.
• Reclaim floor (Problem B). RemoteLogManager takes a logStartOffsetOverride (control-plane cross-tier earliest) so a rebuilt leader no longer reclaims the entire remote classic prefix [earliest, seal).
• Fail-safe reclaim floor. When the override is unavailable (control plane down / metadata not yet propagated), the RLM no longer falls back to the local seal for a consolidating partition — it uses the remote earliest and defers, so a transient control-plane outage can never trigger irreversible over-deletion. A new isConsolidatingDisklessPartition predicate distinguishes this from classic topics, which keep the upstream local-log-start floor.

Docs: DISKLESS_CONSOLIDATION.md updated; full design/rationale and an uncertainties/risk-tracking checklist in DELETE_RECORDS_DISKLESS_ROUTING_IMPL.md.

• RemoteLogManagerTest — reclaim floor uses the override; fail-safe skips reclaim when the override is absent for consolidating; classic topics unchanged.
• ReplicaManagerInklessTest — crossTierEarliestOffset branches (cache hit, miss→CP, CP error/throw→empty, non-consolidating→empty); isConsolidatingDisklessPartition.
• DisklessLeaderEndPointTest — whole-log start = cross-tier earliest; reverts to local seal when unavailable.
• DisklessFetchOffsetRouterTest, DisklessDeleteRecordsForwarderTest, DeleteRecordsCommandTest.
• e2e consolidation_delete_records_across_tiers_test.py — earliest advances off 0, is identical on every broker, settles exactly at the delete boundary, and the surviving prefix reads back contiguous.
• core/storage compile + checkstyle + spotbugs green.
• Open items tracked in the checklist in DELETE_RECORDS_DISKLESS_ROUTING_IMPL.md (multi-partition/rapid-failover e2e, retention×DeleteRecords, sync-DB/batching, etc.).

Delete this text and replace it with a detailed description of your change. The
PR title and body will become the squashed commit message.

If you would like to tag individuals, add some commentary, upload images, or
include other supplemental information that should not be part of the eventual
commit message, please use a separate comment.

If applicable, please include a summary of the testing strategy (including
rationale) for the proposed change. Unit and/or integration tests are expected
for any behavior change and system tests should be considered for larger
changes.