No status_source parallel to text_source

Approved facts record text_source (the planner's original text) and edited_by_human whenever the reviewer edits text. Status edits get no equivalent — status_history records only the reviewer's final value; the planner's proposed_status is dropped.

We keep this asymmetric. Text edits track corrections to a literal quotation that can be objectively wrong (mishearings, typos, ASR errors); preserving the original is part of the claim's truthfulness story. Status is a reviewer judgment call — the planner's guess is a model artefact with no lasting truth value, the same reason plan.yaml itself isn't archived after extraction.

If we ever need "how often does the reviewer override the planner's status?" for model-quality analysis, add it then. Until then, recording it durably suggests it has lasting value, which it doesn't.