Support more granular splitting in BigQueryStorageStreamSource

The BigQuery storage API supports partitioning of streams in the API. However, the partitioning granularity is not sufficient for some pipelines, for which it is desirable to be able to dynamically rebalance down to the level of individual rows. This can be implemented by storing both the stream ID and an offset range as components of a stream source, as opposed to a stream ID only.

Imported from Jira [BEAM-11215](https://issues.apache.org/jira/browse/BEAM-11215). Original Jira may contain additional context.
Reported by: kjung520.