I disagree. They were using it as sample data, when they should have used synthetic sample data. It's totally sane to check-in a JSON (or similar) file with a few hundred [edit: non-confidential] samples so that you can run integration tests without needing to set anything else up.

Similarly, I keep small pdf manuals in git. I add them in their own commit, which doesn't have a useful diff. In exchange, they're always there and I don't need to spin up some special one-off system.