feat: SLA Risk Score (M118)

[hlin99]

Summary

Composite 0-100 SLA risk score combining five signal dimensions into a single actionable metric.

Changes

• SLARiskScorer class in sla_risk.py
• Pydantic models: RiskScore, RiskFactor, SLARiskReport, RiskLevel
• 5 risk factors with configurable weights:
  • Headroom tightness (30%): how close P95 latencies are to SLA thresholds
  • Tail heaviness (20%): P99/P50 ratio indicating long-tail risk
  • Latency jitter (20%): CV of latency indicating instability
  • Convergence adequacy (15%): whether sample size is sufficient
  • Error budget burn rate (15%): rate of SLA-violating requests
• Risk levels: LOW (0-25), MODERATE (25-50), HIGH (50-75), CRITICAL (75-100)
• CLI sla-risk subcommand with table + JSON output
• Programmatic assess_sla_risk() API
• 30 new tests

Closes #259

[hlin99-Review-Bot]

✅ LGTM. Clean implementation — 5-factor composite risk scoring with well-bounded outputs, good test coverage (30 tests), proper edge case handling. CI all green.

[hlin99-Review-BotX]

✅ Approved. 5-factor composite risk scorer is well-designed — clean scoring functions with proper clamping, good Pydantic models, CLI with table+JSON output, and 30 tests. CI all green. Ship it.

[hlin99-Review-BotX]

✅ Approved (hlin99-Review-BotX)

Idea Value: Strong — composite risk scoring combining headroom, tail, jitter, convergence, and burn rate into a single actionable metric is genuinely useful for SLA management.

Code Quality:

• Clean Pydantic models with proper field descriptions
• Scoring functions are well-bounded (0-100) with sensible thresholds
• Convergence check via running P95 windows is a nice touch
• CLI with both table and JSON output
• 30 tests, CI green across Python 3.10/3.11/3.12
• docs/iterations/current.md updated

LGTM.