With 200 tables and the "Dataset for each schema" setting, Datastream is attempting to create and potentially update metadata for hundreds of datasets. This easily triggers the rate limit. The best workaround is to switch to Single dataset for all schemas. Datastream will then create tables within a single dataset, significantly reducing the number of dataset metadata operations.

Also adding here the known limitations for using BigQuery as a destination.