[Promtail] Duplicate prometheus register call when using Kafka

[taisho6339]

Describe the bug
When we use Kafka target in Promtail, we can't use metrics stage due to the /metrics call failure with the following error.

14 error(s) occurred:
* collected metric "my_http_request_duration" { label:<name:"service" value:"nova-api" > label:<name:"env" value:"dev" > histogram:<sample_count:3151 sample_sum:81.74437090000009 bucket:<cumulative_count:3051 upper_bound:0.1 > bucket:<cumulative_count:3130 upper_bound:0.5 > bucket:<cumulative_count:3130 upper_bound:1 > bucket:<cumulative_count:3139 upper_bound:2.5 > bucket:<cumulative_count:3151 upper_bound:5 > > } was collected before with the same name and label values

In kafka target, this ConsumeClaim function is called in every kafka partition.
https://github.com/grafana/loki/blob/main/clients/pkg/promtail/targets/kafka/consumer.go#L86

It causes the duplicate Register call as the following.

(c *consumer) func ConsumeClaim -> (t *targetSyncer) NewTarget -> stages.NewPipeline -> stages.New -> newMetrics

https://github.com/grafana/loki/blob/main/clients/pkg/logentry/stages/metrics.go#L127

To Reproduce
Steps to reproduce the behavior:

1. Started Promtail 2.4.1 with the config like this.

scrape_configs:
  - job_name: dev-kafka
    kafka:
      use_incoming_timestamp: true
      version: 2.4.2
      brokers:
        - xxxx
      group_id: promtail-dev
      topics:
        - grafana-loki-dev
      labels:
        env: dev
    pipeline_stages:
      - match:
          selector: '{env="dev"} |= "access_log"'
          stages:
            - json:
                expressions:
                  request_time:
            - metrics:
                http_request_duration:
                  type: Histogram
                  prefix: my_
                  source: request_time
                  description: "http request duration in API"
                  max_idle_duration: 24h
                  config:
                    action: inc
                    buckets:
                      - 0.1
                      - 0.5
                      - 1.0
                      - 2.5
                      - 5.0

2. call /metrics endpoint

Expected behavior
A clear and concise description of what you expected to happen.

• /metrics call success

[taisho6339]

Let me fix this.