ops: suppress non-blocking LiteLLM P2028 alerts

[seonghobae]

Summary

Adds an admin-scoped operational alert classifier so the known LiteLLM Prisma P2028 spend-log transaction timeout does not become a service outage/page event when it is explicitly non-blocking.

Closes #51.

What changed

• Added classify_operational_alert() with a complete suppression signature for LiteLLM + Prisma + P2028 + transaction-start timeout + [Non-Blocking] + update spend logs.
• Added POST /api/v1/operational_alert_classifications as an admin-only endpoint that returns a derived routing decision without echoing the raw alert payload.
• Kept incomplete or user-facing P2028-like alerts on normal incident policy.
• Documented alert-pipeline wiring and Project Complete architecture-mapped governance and orchestration surfaces #1 coordination per docs: cross-agent protocol for operating the GitHub Project (roadmap source of truth) .github#363.

Verification

• python tests/test_operational_alerts.py
• python tests/test_api_contract.py
• python -m compileall contextual_orchestrator tests/test_operational_alerts.py
• git diff --check
• python -m pytest -q -> 153 passed

Notes

This does not mutate LiteLLM, PgCat, or PostgreSQL configuration. It prevents this specific non-blocking persistence-pressure signature from entering outage/page routing while leaving inference health, latency, 5xx, and customer-facing SLO alerts active.

[opencode-agent]

OpenCode Review Overview

• Head SHA: b6d73e5ca20ccb0183c9fa3561532df4939897fd
• Workflow run: 29060639813
• Workflow attempt: 1
• Gate result: APPROVE (exit 0)

Changed-File Evidence Map

flowchart LR
  PR["PR changed files"] --> Evidence["OpenCode bounded evidence"]
  Evidence --> S1["Changed file (5 files)"]
  S1 --> I1["repository behavior"]
  I1 --> R1["Review risk: Changed file (5 files)"]
  R1 --> V1["required checks"]
  Evidence --> S2["Docs (2 files)"]
  S2 --> I2["operator or user guidance"]
  I2 --> R2["Review risk: Docs (2 files)"]
  R2 --> V2["docs review"]
  Evidence --> S3["Test: test_operational_alerts.py"]
  S3 --> I3["regression suite"]
  I3 --> R3["Review risk: Test: test_operational_alerts.py"]
  R3 --> V3["targeted test run"]

Loading

[seonghobae]

Closing this branch because it modeled the problem as alert suppression. CodeGraph review shows the correct fix belongs in PR #46: usage/cost telemetry must be non-blocking and prompt-safe at the contextual-orchestrator boundary.