A telemetry application writes millions of independent device readings to DynamoDB. One table uses the constant value all-devices as its partition key and a timestamp as its sort key. Writes are throttled despite substantial table capacity. Queries normally retrieve one device's readings by time range. Which redesign best addresses the hot-key pattern?

1. Use one global secondary index with the same constant partition key for all records.
2. Move every write to one new constant partition key named current-month.
3. Keep the constant partition key and add a new sort-key prefix for every month.
4. Use the device identifier as the partition key and a time-based sort key, validating distribution for the expected device traffic.