feat: stat_middleware

[yudytskiy]

Prometheus is good, but it uses file operations on each middleware hook; that might become a bottleneck.

Suggested middleware uses similar metrics, but it keeps statistics in memory and may be queried by the simple task when needed.

As soon as a middleware instance starts within a worker process, requesting stats requires getting stats from each worker middleware instance.

I suggest starting an additional worker task inside middleware for a special pub-sub stat broker, so kiqing its task leads to execution on each main worker process, gathering all results together.

This pull request is just an idea implementation and includes:

• metrics classes
• middleware using metrics
• test for metrics
• demo example stats script

I had to bump up the required mypy and black minimal versions to support modern generic syntax.

So I just share the idea of request-oriented statistics gathering via pub-sup broker.