Refactor BidiWriteObject checkpoint and data flush to better follow GCS server

[cojenco]

Details will be commented in internal issue

[cojenco]

          Just an idea (no action needed): maybe we should let the client pick the behavior (via headers like we do for retry testing). That way can write tests for different scenarios (e.g. when the data is not flushed and we disconnect, or when the data is flushed before the client expects it to).

Originally posted by @coryan in #586 (comment)

^^
Additionally, explore testing scenarios around checkpoint and data flush