In https://github.com/mongodb/mongo-ruby-driver/pull/2026, bulk write split of requests where each document was under 16m, the total bulk write size was over 16m and the total bulk write compressed to under 16m was fixed to work (because the server rejects bulk write which when expanded exceeds 16m).

However due to our current operations layer API, we redundantly serialize the message at the finally accepted bulk write size. We should look into refactoring our API to allow compressing a message that is already serialized to bson, vs the current behavior of compressed message requiring the original uncompressed message.

If not handled in https://jira.mongodb.org/browse/RUBY-2241, we should also verify that the bulk writes are split based on document sizes and not based on message sizes, since message sizes contain the 16k allowance and we would like to report to users that they exceed document size limit at 16m+1 bytes.