Improve inserter delivery guarantees

I am using the [inserter](https://docs.rs/clickhouse/latest/clickhouse/inserter/struct.Inserter.html) to periodically write batches of messages from a Kafka topic into ClickHouse. I wonder what happens to the data when an error happens on write/commit. It seems like the internal buffer is dropped and the unfinished binary data stream is aborted.
This essentially means, that any data is lost on error. Is this true? What are some best practices to guarantee data delivery or how would I best implement a retry mechanism?

Start with the linked Inserter documentation and trace the write/commit behavior described in the issue. Determine whether buffered data is lost when a write or commit fails, then document the observed delivery guarantees and retry guidance for batches read from Kafka and written to ClickHouse.

Written by the indexing model from the issue text.