Several independent Kinesis consumers read the same stream and compete for shared GetRecords read throughput. Each consumer needs low-latency access to the complete event stream without dividing records among applications. Which feature is the most direct fit?

1. Use a single shared polling consumer and discard events after the first application processes them.
2. Increase only the producer batch size and leave all consumer read configurations unchanged.
3. Register the applications as enhanced fan-out consumers and use their dedicated read-throughput path.
4. Put all applications in one consumer group that assigns each record to only one application.