feat: track out of budget, rate limits, message too large errors explicitly

[subpath]

Jira ticket: https://mozilla-hub.atlassian.net/browse/GENAI-3942
Dashboard: https://yardstick.mozilla.org/d/aonbrtc/firefox-ai-executive-dashboard-server-side?folderUid=ff4lo7wllzldsb&orgId=1&from=now-7d&to=now&timezone=browser&var-query0=&var-environment=prod&var-query0-2=&var-datasource=adpvtjmrxoc1sb&showCategory=Histogram

What's new:

Prometheus metric to track completion rejections with service-type, model-name and rejection errors

QA:

• unit tests old and new ✅
• Local QA:
  Metrics after the Out Of Budget event:

.....
# TYPE mlpa_chat_completions_with_tools_total counter
# HELP mlpa_chat_tool_calls_per_completion Distribution of tool calls per completion.
# TYPE mlpa_chat_tool_calls_per_completion histogram
# HELP mlpa_chat_requests_with_tools_total Number of chat requests that included a tools payload.
# TYPE mlpa_chat_requests_with_tools_total counter
# HELP mlpa_chat_request_rejections_total Number of chat requests rejected due to budget, rate limit, or payload size.
# TYPE mlpa_chat_request_rejections_total counter
mlpa_chat_request_rejections_total{model="openai/gpt-4o",reason="budget_exceeded",service_type="ai"} 1.0
# HELP mlpa_chat_request_rejections_created Number of chat requests rejected due to budget, rate limit, or payload size.
# TYPE mlpa_chat_request_rejections_created gauge
mlpa_chat_request_rejections_created{model="openai/gpt-4o",reason="budget_exceeded",service_type="ai"} 1.773301188783722e+09