Once all the machinery is in place, this ticket is about figuring out the place to hook the checks for retryability against the transactions collection and ensure write idempotency on mongod.

It also includes figuring out whether some form of caching is necessary and taking care of writes which might still be in progress.

Also need to consider how to handle a single write command being split into multiple batches by mongos (a single shard can potentially receive more than one "mini batch" per write command)