Replace pickle with json serialization

[kostaleonard]

Pickle has known arbitrary code execution vulnerabilities. These vulnerabilities are mitigated by the fact that we are only unpickling objects from local files and S3 buckets that the user trusts; however, there is still a risk that the user inadvertently unpickles a malicious object that they did not produce themselves. Changing the serialization method from pickle to json would provide a higher level of security to users.

[kostaleonard]

This is a tough issue. The whole purpose of pickling the data processor object is so that you can execute arbitrary code; there could be a potential workaround to preserve the data processor state in such a way that future examples are preprocessed precisely according to the data processor standards at the time of versioned dataset publication (not necessarily in line with any git commit).