In working on NODE-1702, I realized that Node's Unordered Bulk Write implementation will only propagate a non-(write|writeconcern) error if it happens to occur on the last batch to return. This is definitely not proper behavior, but I could not find in our specification what the proper behavior is in this case.

Some example cases:

1. A Network Error that affects only one of the batches in a bulk write. All other writes succeed
2. An Auth error that happens on all the batches
3. A Server-Side error (malformed query args, BSON max size exceeded, etc.)

What is the correct way for an unordered bulk write to handle an error in these cases?