ref(producer): minor metrics improvements

[bmckerry]

Currently, taskproducer metrics look strange:

• latency max spikes up to 1s: https://app.datadoghq.com/s/FH6-Y3/ij5-abd-evg
  • I think the latency issue is because when the worker tries to fetch from an empty queue, it has a 1s timeout before it runs check_task_future_completion()
  • This PR tries to fix this issue by having the worker run check_task_future_completion() before it tries to fetch a new task
  • This also simplifies the logic by only calling check_task_future_completion() from a single place
• very few activations with pending futures are reported: https://app.datadoghq.com/s/FH6-Y3/cpk-wy4-7mv
  • I think this is because the majority of tasks don't produce anything, so any gauge samples that show a value are getting drowned out by samples reporting 0
  • This PR tries to fix this by only having the gauge report when there are pending activations in the queue