Detect corrupt data at the PR stage

[holly-cummins]

https://github.com/quarkusio/benchmarks/tree/main/results/spring-quarkus-perf-comparison/2025-11-24_20-47-45 is missing data because of a problem on the run. We don't want to overwrite the latest json data sets with bad data. We could either:

• Add a CI job which runs a sense check (just file length, even?) on the data PRs to block auto-merging if the data is bad
• If we do want the logs, we could adjust something to merge the run data, but not update latest. That would maybe be a two-stage thing where it merges, and then another job runs and updates latest.

Ideally both of those checks would also happen before even creating the PR, but I think it's helpful to have two sets of gates, one on each side.