Metrics are not multiprocess

[alexppg]

Describe the bug
Is not a bug per se, since we don't have to change anything in pyms to make it work, what #179 did was ok. The problem is that it's not enough.

The problem can be reproduced using pyms with gunicorn with at least two workers and do a lot of http requests to the metrics endpoint. You'll see that each response has different values, which shows that each worker has it's own count of the metrics.

I'll add this just to track it, I'm seeing at he moment how to fix it. It will probable be just a PR with more documentation.