As my priority is for now not the indexer, i don't have time to find the proper way to fix the mimetype indexer (more information in the parent task T861).
So taking a shortcut to unstuck the indexers to avoid having them stuck in a rescheduling loop hell.

The idea is to trap the error raised.
Log in error the hash and continue indexing the rest.
A pattern like Detecting mimetype error for id <hash> will help.
This way, we can compulse a list of sha1 to re-index later on when the fix is found and deployed.