allDocs ndjson response

Would be nide if the _all_docs endpoint could respond in http://ndjson.org/ format (similar to _changes) which would make it possible to do streaming parsing of the response which can improve memory usage and also parsing performance of clients.

Start at the _all_docs HTTP endpoint and compare its response handling with the existing _changes endpoint. Determine how an ndjson response should be exposed without breaking current clients, then verify that _all_docs can be consumed incrementally with reduced buffering and parsing overhead.

Written by the indexing model from the issue text.