The reason for the problem UPDATE/MERGE must match at most one source row for each target row is that your INNER JOIN in the first query tries to update rows in documents_bkup when there are several matching rows in transactions_bkup depending on transaction_header_guid. Because an INNER JOIN returns all matching rows from both tables, the update statement becomes unclear which case_guid to use if a transaction_header_guid from documents_bkup appears more than once in transactions_bkup with various case_guid values. Due this ambiguity, bigquery throws an error. Refer to this documentation for more information.

Using aggregate functions in the JOIN clause to choose a single case_guid for each transaction_header_guid may be necessary to use the INNER JOIN successfully (e.g., MIN(case_guid), MAX(case_guid)) And also refer this stack link1, link2.