Add Phase 9 orchestration evaluation quality gates

[paullizer]

Summary

Implements Phase 9 of chat orchestration as an evaluation and quality-gate slice on top of the merged Phase 8A/8B contracts.

• Adds six deterministic golden scenarios covering M365 profile plus selected image, SQL metrics, missing public-profile evidence, selected-image Q&A, current Fairfax rules, and governed-agent evidence-only central synthesis.
• Adds fixed-schema privacy-safe evaluation events for terminal run/source outcomes, recommendation creation/decision/outcome, latency, citation yield, missing evidence, and unsupported-fact counts.
• Adds a repeatable repository-local quality-gate runner and an opt-in five-scenario controlled live-smoke manifest/result contract.
• Repairs orchestration-adjacent stale version, Blueprint, helper-fixture, and false-green test assertions.
• Updates the application version, feature documentation, and release notes to 0.250.068.

Privacy And Governance

• Hashes run correlation values and reports discovered agents only as governed_agent.
• Allowlists capability classes, reason codes, statuses, and task profiles.
• Excludes prompts, evidence/response text, canonical agent and object IDs, private scope details, inaccessible counts, endpoints, secrets, hidden tools, action names, and action arguments from evaluation events.
• Leaves existing authorization, exact source-turn revalidation, ETag decision/resume idempotency, process-restart reconciliation, and central finalization behavior unchanged.

Validation

• 203 passed, 5 skipped using scripts/run_phase9_orchestration_quality_gates.py.
• The five skips are explicitly environment-gated browser/live scenarios.
• All changed and new Python files passed py_compile.
• Full-file Broken Access Control guardrail passed for all 4 changed application Python files.
• Full-file XSS sink guardrail passed for all 4 changed application Python files.
• Route-policy suites, diagnostics, and tracked/untracked whitespace checks passed.
• Controlled live smoke was not executed because no explicit live manifest/auth configuration was supplied; its local contract test passed and live execution remains opt-in.

Scope Boundary

Phase 10 remains responsible for generalized output types, broader multi-agent execution, Foundry or action-attached discovery, consequential/write approval, durable in-flight execution, and complex workflows.

Refs #1021