feat(quality): add AI-assisted OSPS-QA-06.02 evaluation

[vinayada1]

Summary

Implements an AI-assisted evaluator for OSPS-QA-06.02 (documentation explains when and how tests are run), replacing the previous hard-coded NeedsReview stub.

• Maps OSPS-QA-06.02 to the new quality.TestExecutionDocumentation evaluator.
• Sends a conservative evidence boundary (README + CONTRIBUTING) to the configured AI provider and parses a strict, schema-validated verdict.
• Persists AI evidence packets via the SDK writer (sdkai.WritePacket), gated behind the SDK's double opt-in (config.Write + ai_write_evidence: true), with SDK-side redaction.
• Caches successful verdicts per run; provider/config/schema failures remain retryable.

Backward compatibility

• AI not configured: behavior is identical to today. When Config is nil, the client is nil, or ai_provider/ai_api_key are missing, the evaluator returns the same gemara.NeedsReview result and message as the previous stub. All failure paths fall back to the same NeedsReview verdict.
• SDK version: requires privateer-sdk v1.28.0 symbols (WritePacket, PacketAttempt, NewSanitizer, RedactPatterns, NewClientFromConfig). go.mod already pins v1.28.0, and Go's minimum version selection guarantees the floor at build time, so there is no silent old-SDK runtime risk.