Compares actual layer data on the compute resource against stored metadata.
Auto-corrects size and file count mismatches, and flags structural issues
(e.g. metadata without data, or data without metadata).
Returns the updated cost summary along with any corrections and issues found.