PR 9994 introduces realtime lag metric but it add a monitoring thread per table: can we use a thread for all tables

PR 9994 introduces realtime lag metrics per partition and create a monitor thread per table to time out metrics for partitions that have been migrated out of the server: opening this to answer the following question: can we use a single thread for this?

Start by reading PR 9994 and its realtime lag metric implementation, focusing on how partition timeouts and per-table monitor threads are managed. Trace the lifecycle across migrated-out partitions and determine whether a shared thread can preserve the same behavior; document the design decision and its trade-offs.

Written by the indexing model from the issue text.