Memory leak in PCMetrics class

My team is using parallel-consumer in production, and we observed increasing memory usage over time in all of our services.
After monitoring the heap, we found that this stack trace was responsible for 96% of the heap size after 3 days of uptime:



This is reproducible locally. Whenever the parallel-consumer attempts to commit its offsets, it creates two new timers and adds them to the `registeredMetrics` list in PCMetrics, even if the exact tag combination already exists.

Keeping track of the `registeredMetrics` list shows this after 10 minutes (With a 1 second commit interval):

Changing the list to a set seems to fix the issue, but not sure if a more sophisticated method would be preferred.