Fix ClickHouse lookup pagination for UNION queries

[CheekyNox]

Summary

• Paginate ClickHouse union lookups globally instead of limiting only one union branch.
• Use a two-phase lookup so deep pages select lightweight row keys before loading binary payload columns.
• Stop processing cleanly when a lookup query fails instead of triggering a secondary null-pointer exception.

Problem

ClickHouse lookups that combine multiple event families append pagination directly after a multi-branch UNION ALL. The trailing LIMIT does not constrain the complete union, so a lookup configured for a few lines can return thousands of rows and flood the Minecraft client chat.

Applying an outer ORDER BY ... LIMIT ... OFFSET ... directly to the full rows fixes the result count but makes deep pages materialize large item/container payload expressions before pagination. On large histories this can exceed the ClickHouse memory limit.

Fix

For paginated ClickHouse detail lookups:

1. Query only tbl, id, and time across the candidate sources and apply global ordering/pagination.
2. Load complete rows only for the selected source/row IDs.

This keeps pagination global while avoiding materialization of binary payloads for discarded rows. Non-ClickHouse lookup paths are unchanged.

Verification

• mvn verify (BUILD SUCCESS; tests are skipped by the upstream Maven configuration)
• Reproduced the original excessive result output with a multi-family ClickHouse lookup
• Verified the initial global-limit fix caps output to the requested page size