Repository with Broken Shards' index-N File Can't Create New Snapshots

[original-brownbear]

Due to some bug that I'll track down here, we were (and maybe still are) writing duplicate JSON keys into snapshot shards' index-N blobs. This leads to those blobs not being read correctly, which makes it impossible to create further snapshots for a shard with such a broken index-N blob.

This was technically triggered by making the JSON parser throw on this kind of data in #22073 most likely.

---

• We need to fix writing broken JSON here if it still happens
• We need to harden the logic here to be able to deal with this kind of broken index-N blob to get repositories out of this broken state.

[elasticmachine]

Pinging @elastic/es-distributed