Ensure sample_id is always populated

[jklukas]

As we continue to have discussions about data retention policies, sampling may become an even more important concern where we permanently delete data after a certain period of time based on sample_id.

We could consider adding more robust support for sampling, with schemas defined in mozilla-pipeline-schemas including metadata about the field(s) to use for calculating sample_id. Absent such a field, we could fall back to per-document sampling rather than per-client by calculating sample_id from document_id.

cc @mreid-moz