feat: add --generic-hostname-worker-task-metric flag

[bafulton]

Problem

The executing worker's hostname label drives task-metric cardinality on autoscaled deployments: celery_task_runtime and the celery_task_* counters are all labeled with the worker's hostname, which in Kubernetes/KEDA setups is an ephemeral pod name. The name × hostname × queue set therefore churns and grows with the worker fleet. --generic-hostname-task-sent-metric solves this on the client side, but there is no equivalent for the worker-executed metrics.

Solution

Add --generic-hostname-worker-task-metric, mirroring the task-sent flag: it labels the celery_task_* counters and celery_task_runtime with a generic hostname instead of the worker's. track_task_event builds a single labels dict that feeds every counter and the runtime observation, so one gated assignment collapses them together. celery_task_sent keeps its own flag so the two compose, and celery_worker_up / celery_worker_tasks_active are untouched since they do not pass through track_task_event.

The test mirrors the existing generic_hostname_task_sent_metric test and passes on the memory, redis, and rabbitmq brokers.

Refs #387

[bafulton]

@danihodovic, I'd love to get your thoughts when you have a moment! I'm currently running into cardinality issues in my prod environment; Celery Exporter is generating more than 90,000 independent metrics and it's breaking prometheus and datadog scrapes.