Metrics have invalid `pid` specification

Currently, some of the metrics use pid as Process.pid, and due to hardcoded implementation assuming that all metrics are gathered for Unicorn.

This changing pid, increases the churn-rate on Prometheus, due to high rate of changing labels and large amount of metrics created by the process.

We should never use Process.pid for metrics, but instead, we should use relative worker identifier (0..n | master).