feat(observability): add Prometheus /metrics endpoint (Phase 10)

[JesseJohn7]

Summary

Adds a Prometheus-compatible /metrics endpoint publishing request, token and cost counters in the text exposition format.

Changes

• New app/observability/ package:
  • metrics.py: small thread-safe counter registry and text exposition renderer (no new dependencies).
  • routes.py: GET /metrics (text/plain; version=0.0.4, Cache-Control: no-store), disabled with 404 when METRICS_ENABLED is false.
  • __init__.py: init_observability(app) registers the route and counts requests in after_request.
• Metrics exposed:
  • http_requests_total{method,endpoint,status}
  • llm_tokens_total{provider,model,direction}
  • llm_cost_usd_total{provider,model}
• Token and cost usage is recorded via record_usage(...) where assistant messages are saved.
• docs/observability.md updated to reflect what is now implemented.

Bounded and fast

• Scrapes only read in-memory counters; no database queries.
• endpoint label is the Flask route pattern (unmatched paths share one unmatched series), and status is grouped (2xx, 4xx).
• Each metric is capped at 500 label combinations; extra series fold into an other label.

Tests

tests/test_metrics.py covers HELP/TYPE lines, sample lines, label escaping, negative increments, the cardinality cap, content type, validity of every line, route-pattern labeling, token/cost counters, and the disabled switch.

ruff check ., black --check . and pytest all pass locally.

Closes #476

[drips-wave]

@JesseJohn7 Great news! 🎉 Based on an automated assessment of this PR, the linked Wave issue(s) no longer count against your application limits.

You can now already apply to more issues while waiting for a review of this PR. Keep up the great work! 🚀

Learn more about application limits