CONFLICT:
- cloud service implementation didn't exist in this version

Updates the filter used when collecting metrics samples.

In some cases, the samples take the form:

Sample(name='redpanda_kafka_max_offset', labels={'redpanda_namespace': 'kafka', 'redpanda_partition': '5', 'redpanda_topic': 'panda-topic'}, value=0.0, ...

(cherry picked from commit e567d63)

Read replicas were previously translating their own local log offsets to
return `redpanda_kafka_max_offset` to the metrics endpoint. This is
different from how read replicas calculate the HWM when returning to the
Kafka endpoint, which just goes directly to cloud storage.

This adds the same read replica check that we have in the Kafka layer.

(cherry picked from commit 7e33cc7)