Then I'm speechless.
Can you reindex?
Old versions were not having checksum so when moving shards or merging it could happen that.
Definitely upgrading and keeping at least with the most recent versions of the major version you are using would have help to avoid that or at least to discover sooner that kind of problem.
May be you could try to start a 5.2 cluster and see if reindex from remote API could help you.
Good luck.