Fix continuous integration builds and make JSON support more robust at build

[tmadlener]

BEGINRELEASENOTES

• Only build JSON support if a suitable version of nlohmann/json is found. This should fix a few issues in CI.
• Fix formatting of edm4hep2json.
• Update versions of github actions used in CI to latest available.

ENDRELEASENOTES

[tmadlener]

key4hep nightlies are failing because of HEP-FCC/FCCAnalyses#227

[tmadlener]

Merging without waiting on the FCCAnalyses fix, in order to fix at least a few of the other CI workflows.

[vvolkl]

Thanks for the fix - something is still wrong though when fccanalyses can break the edm4hep tests, that's not how dependencies should work... Maybe a name collision in the rdf test?

[tmadlener]

Yeah, I agree that this really should not happen, but I suspect the culprit in this case is ROOT that loads dictionaries a bit too eagerly. I suppose it would be possible to set the test environment even stricter to exclude, e.g. other dictionaries to be discoverable during tests. However, I seem to remember, that I have struggled with this already quite a bit in the past, due to how different e.g. an LCG (view) environment is compared to one from a key4hep stack build.