DocumentInserter drops whole task batches when a payload key contains a dot

Hit this capturing real agent telemetry: a task whose payload embedded a CSV profile with a column literally named yr.doy. Mongo rejects . (and leading $) in field paths, and since mongodb_dao.insert_and_update_many_tasks writes tasks via $set upserts in a bulk_write, the whole batch raises BulkWriteError (code 16412) and every task in it is lost:

pymongo.errors.BulkWriteError: ... 'code': 16412,
'errmsg': "Invalid $set :: caused by :: FieldPath field names may not contain '.'.
           Consider using $getField or $setField."

Could we sanitize/escape keys before building the update ops? Or is the expectation that the user does it? We currently work around it producer-side by escaping ./$ in payload keys before emission.

Start at mongodb_dao.insert_and_update_many_tasks, where payload keys become $set upserts in a bulk_write; reproduce the failure with a key such as yr.doy and inspect the resulting BulkWriteError. Establish the expected key encoding for MongoDB, then verify that a batch containing dotted or leading-$ keys completes without dropping tasks.