read replica partitions report incorrect kafka max offset

[andrwng]

Version & Environment

Redpanda version: (use rpk version): dev

for metrics:

redpanda/src/v/cluster/partition_probe.cc

Lines 208 to 222 in 1e8887e

	sm::make_gauge(
	"max_offset",
	[this] {
	auto log_offset = _partition.high_watermark();
	auto translator = _partition.get_offset_translator_state();
	try {
	return translator->from_log_offset(log_offset);
	} catch (std::runtime_error& e) {
	// Offset translation will throw if nothing was committed
	// to the partition or if the offset is outside the
	// translation range for any other reason.
	return model::offset(-1);
	}
	},

vs for kafka endpoint:

redpanda/src/v/kafka/server/replicated_partition.h

Lines 78 to 87 in 1e8887e

	model::offset high_watermark() const final {
	if (_partition->is_read_replica_mode_enabled()) {
	if (_partition->cloud_data_available()) {
	return _partition->next_cloud_offset();
	} else {
	return model::offset(0);
	}
	}
	return _translator->from_log_offset(_partition->high_watermark());
	}

This branch to determine whether we're a read replica should take place for both the Kafka endpoint and the metrics endpoint, seeing as how users may want to track max offset via metrics. The latter is not happening today.