feat: add native Prometheus metrics exporter to stats server

[lavr]

Motivation

Prometheus has become the de facto standard for application monitoring. In practice, virtually no production deployment goes without Prometheus metrics - it is the first thing teams set up for observability.

In real-world production environments, it is essential to monitor uWSGI internals: worker busyness, request rates, memory consumption, harakiri counts, queue lengths. Currently, the only way to expose these metrics to Prometheus is to run a sidecar exporter (typically uwsgi_exporter). This adds operational complexity - an extra container/process to deploy, configure, and maintain - and consumes additional resources, all just to translate JSON stats into Prometheus format.

This PR eliminates the need for a sidecar by adding native Prometheus exposition format support directly into the uWSGI stats server.

Changes

• GET /metrics on the stats server returns Prometheus text exposition format
• All other paths return JSON as before - fully backwards compatible
• Metric names are compatible with timonwong/uwsgi_exporter for drop-in replacement
• The prometheus path is configurable via --stats-prometheus-path
• The HTTP stats layer is refactored into uwsgi_stats_read_request() / uwsgi_stats_send_http_header() with enum uwsgi_stats_format for clean format selection
• Integration tests for both JSON and Prometheus endpoints

Usage

No additional flags required in uwsgi config. The /metrics endpoint is automatically available on the existing stats HTTP server:

[uwsgi]
stats = :9191
stats-http = true

Prometheus scrape config:

scrape_configs:
- job_name: uwsgi
  static_configs:
    - targets: ['localhost:9191']

No extra flags, no sidecar.

Closes #1733