fix(store): reject cross-shard duplicates during standby restore

[982945902]

Summary

• reject standby snapshot entries when the same (tenant, key) already exists under the current metadata routing
• perform the duplicate check before restoring metrics, segment state, or routing metadata
• add regression coverage for ungrouped-to-grouped and grouped-to-different-group cross-shard duplicates

Root cause

RestoreFromStandbySnapshot() checked only the shard selected by the incoming snapshot entry's group_id. If an existing object was routed by its key or by another group, restore could insert duplicate metadata into a different shard.

Tests

• MasterServiceHATest.RestoreRejectsUngroupedObjectDuplicatedIntoAnotherShard
• MasterServiceHATest.RestoreRejectsGroupedObjectDuplicatedIntoAnotherGroupShard
• all MasterServiceHATest.Restore* tests: 11 passed
• full master_service_ha_test: 83 passed, 1 unrelated failure (MasterServiceBatchRecordE2ETest.PromotionCatchesUpToDurablePrefix) because the local build used STORE_USE_ETCD=OFF; tracked by [Bug]: HotStandbyService tests do not exercise claimed etcd behaviors #3496
• git diff --check

Fixes #3398

[Icedcoco]

LGTM