Aggregator recover fail when base_table is older than aggr_table

**Describe the feature you'd like**

When the table has multiple replicas, when one fails, the leader may change. New leaders of the base_table and aggr_table may be different, causing the base_table may be older than the aggr_table.

We have to invalid the affected entries in the aggr_table, and re-compute to make it right.

**Additional context**

long-window support

Look at the aggregator recovery logic in the codebase, focusing on how base_table and aggr_table replicas are managed during leader changes. Identify where invalidations and recomputations should happen, and examine any existing tests for long-window support or replica failure scenarios.

Rédigé par le modèle d'indexation à partir du texte de l'issue.