PR #169 made document comparison use a memory pool. That changeset neglected to update uses of bson.Raw.Lookup(), which does not clone the field’s raw value. Thus, when a mismatched document was “put back” into memory, another thread overwrote a part of it.

Thus we got mismatch documents (i.e., in verifier metadata) where the same document ID showed up twice. This also, critically, made the verifier skip necessary rechecks because a later mismatch overwrote the buffer that contains the document ID meant for recheck.

This changeset fixes that problem by always cloning the document ID as part of reporting a mismatch. It also beefs up testing to ensure that the expected document IDs are reported in the log and in the rechecks.

This also makes the field-order-difference mismatch text more concise by removing a redundant document ID.