It doesn't invoke new ShardId(...) which is where the exception was,
but it's still nice to report this information if we have it (which we
do!).

Failure is so much more normal/expected than dropping, so let's have
that one first. Also it has a better explanation of what's going on
and why, which I like having earlier (but don't want to duplicate).

Increasing the symmetry between markItemAsFailed and
markItemAsDropped.

This reverts commit 29ce3c6.

This requires that we have a default_pipeline in place, so I've also
extended the cleanup to include deleting the index -- otherwise if
something goes wrong the cleanup can fail because you can't drop a
pipeline if there's an index using it.

The response will be an UpdateResponse (not an IndexResponse) when a
document is dropped on an upsert, both of those are a
DocWriteResponse.

This requires that we have a default_pipeline in place, so I've also
extended the cleanup to include deleting the index -- otherwise if
something goes wrong the cleanup can fail because you can't drop a
pipeline if there's an index using it.