The current default, 0.8.2, is a 5 year-old version. The problem with defaulting to older versions when the Kafka cluster is a newer version is manifold:

- requires Kafka to downconvert consumer response messages (e.g if a newly-configured client produced messages and a default Sarama client is reading) - this has implications on the performance of Kafka, because it uses additional CPU. Especially so if compression is enabled, at which point the broker would need to decompress, downconvert and compress the message back. Downconversion also blocks zerocopy, because your fetch response now needs to be copied into memory
- requires Kafka to upconvert producer requests' messages
- in general is hard to support. While the latest Kafka currently supports the oldest versions, it is in the best interest of the project to deprecate and eventually drop support for legacy versions. Otherwise it becomes hard to maintain, the test matrix grows and new features need to work around old version limitations (no idempotency, exactly once). It is easier for Kafka to deprecate/drop support when its ecosystem has done so already

0.11 is the minimum we should default at as it enables Kafka's v2 message format, avoiding expensive upconversion/downconversion in the Kafka broker. It is also required for correctness in some cases (e.g KIP-101)
1.0.0 is a 3-year old version at this point and a reasonable default

This reverts commit 91e853d.