Zoekt: Zero-downtime reindexing creates too many tasks

Summary

After bumping up the schema version in Zoekt v1.2.0, we encountered a problem on staging which caused 2025-08-08: GSTG: Zoekt creates tasks in a loop (gitlab-com/gl-infra/production#20322 - closed)

Steps to reproduce

What is the current bug behavior?

What is the expected correct behavior?

Relevant logs and/or screenshots

Possible fixes