ce_ prefix is no longer applied automatically for Cloud Events sent to Kafka

[DidierLoiseau]

Describe the bug
Since 5.0, following #1294 (and spring-cloud/spring-cloud-stream#2222), it appears that Spring Cloud Function no longer determines automatically the correct ce_ Cloud Event prefix to use when a message is sent to Kafka. It only works properly when at least one of the headers has the kafka_ prefix (e.g. when using constants from org.springframework.kafka.support.KafkaHeaders), but if no such header is used, it falls back on the default ce- prefix.

Sample
This can be reproduced easily with function-sample-cloudevent-stream’s DemoApplicationTests.

According to the README (and also this Spring blog post which I think is the same):

Also note that on the receiving end Cloud Events attributes are now prefixed with ce_ prefix (see Cloud Events Kafka protocol bindings), since it was determined by the framework that the target destination is Apache Kafka.

However, when I run that test, I see it uses the cloudevents: prefix instead:

Result Message: GenericMessage [payload={"person":{"firstName":"John","lastName":"Doe"},"id":884,"message":"Employee 884 was hired on 04-09-2026"}, headers={cloudEvents:id=e790a03d-1033-4a9a-a69b-f40727209fc1, cloudEvents:type=io.spring.cloudevent.Employee, kafka_timestampType=CREATE_TIME, kafka_receivedTopic=hire-out-0, kafka_offset=2, cloudEvents:specversion=1.0, message-type=cloudevent, kafka_consumer=org.springframework.kafka.core.DefaultKafkaConsumerFactory$ExtendedKafkaConsumer@270ab7bc, cloudEvents:source=http://spring.io/, kafka_receivedPartitionId=0, contentType=application/json, kafka_receivedTimestamp=1788535901863, kafka_groupId=test}]

and also from Kafkbat UI, I can see that the Kafka headers use the wrong prefix:

{
	"cloudEvents:id": "1c75b58c-b653-4713-a921-258d02ad331d",
	"spring_json_header_types": "{\"cloudEvents:id\":\"java.lang.String\",\"cloudEvents:type\":\"java.lang.String\",\"cloudEvents:specversion\":\"java.lang.String\",\"message-type\":\"java.lang.String\",\"cloudEvents:source\":\"java.lang.String\",\"contentType\":\"java.lang.String\"}",
	"cloudEvents:type": "io.spring.cloudevent.Employee",
	"cloudEvents:specversion": "1.0",
	"message-type": "cloudevent",
	"cloudEvents:source": "http://spring.io/",
	"contentType": "application/json"
}

As a workaround, when using CloudEventMessageBuilder it should be possible to explicitly add a kafka_ header, or to call build() with the right prefix, but then the code is no longer agnostic vs. the destination.

When I read spring-cloud/spring-cloud-stream#2222, it is unclear whether this workaround is the intended solution.

p.s. there are no release notes for 5.0.0?