fix(logger): Improve error handling to return errors when creating Kafka topics (#640)

# Description
It appears that calling `CreateTopics`
[here](https://github.com/caraml-dev/merlin/blob/eb52b7f240e3cc516026816e15b3b248a1aa3850/api/pkg/inference-logger/logger/kafka_sink.go#L59)
in `NewKafkaSink` sometimes doesn't return errors within the `err`
return value, but instead in each of the individual `topicResults` (this
is a slice) elements returned.

This happens in certain scenarios, e.g. when there are authorisation
problems creating new topics, which causes errors to be returned within
each element of the `topicResults` slice instead (leaving `err` as `nil`
instead):

![Screenshot 2025-04-22 at 15 19
52](https://github.com/user-attachments/assets/d9f828a6-2602-4a05-a5bc-491a43dd4825)

This is problematic because it means that even though there were
problems creating the topic, a `nil` error gets returned, with a `nil`
sink value.

Tracing this back to where `NewKafkaSink` gets called in `getLogSink`,
https://github.com/caraml-dev/merlin/blob/main/api/cmd/inference-logger/main.go#L362,
and to where `getLogSink` gets called,
https://github.com/caraml-dev/merlin/blob/main/api/cmd/inference-logger/main.go#L130,
it's easy to see that the error isn't properly caught (since the error
returned is `nil`) at any part of the call stack.

To make things worse, because `logSink` is `nil`, it continues getting
passed around the entire logger code as if it were non-nil, creating
huge problems downstream, like null pointer exceptions (and crashes)
because `logSink` is not at all expected to be `nil`.

This PR simply refactors the error handling of the Kafka sink
initialisation step, so that any errors are properly returned and
subsequently caught by the upstream callers.

# Modifications
- `api/pkg/inference-logger/logger/kafka_sink.go` - Improve error
handling

# Tests
<!-- Besides the existing / updated automated tests, what specific
scenarios should be tested? Consider the backward compatibility of the
changes, whether corner cases are covered, etc. Please describe the
tests and check the ones that have been completed. Eg:
- [x] Deploying new and existing standard models
- [ ] Deploying PyFunc models
-->

# Checklist
- [x] Added PR label
- [ ] Added unit test, integration, and/or e2e tests
- [x] Tested locally
- [ ] Updated documentation
- [ ] Update Swagger spec if the PR introduce API changes
- [ ] Regenerated Golang and Python client if the PR introduces API
changes

# Release Notes
```release-note
None
```

Author: deadlycoconuts

api/pkg/inference-logger/logger/kafka_sink.go

@@ -63,9 +63,12 @@ func NewKafkaSink(
 			ReplicationFactor: ReplicationFactor,
 		},
 	})
+	if err != nil {
+		return nil, err
+	}
 	for _, result := range topicResults {
 		if result.Error.Code() != kafka.ErrNoError && result.Error.Code() != kafka.ErrTopicAlreadyExists {
-			return nil, err
+			return nil, fmt.Errorf("failed to create topic %s: %w", result.Topic, result.Error)
 		}
 	}