Caches are based on the current ref, or a parent ref / base branch, as well as key.
When we update nanopb we'll need to update the cache key too, it's obvious as there are a few lines mentioning the version number.

This has now run once, causing a cached item to appear under the ref/pulls/this-one cache, and the second run has used that cached item for all the jobs.
Once we get it into main then one run will generate a cached item in the main cache which other things can utilise.