feat(fableloom): add editorial autopilot and playthrough QA

[atomantic]

Summary

• add a single series-level AI evaluate/remediate action with explicit provider, model, and effort controls
• add a bounded editor/reviewer autopilot that iterates on deterministic diagnostics and narrative-quality verdicts
• add an exhaustive bounded playthrough harness with transition coverage, termination checks, and optional whole-series AI review
• keep episode-outline readiness honest across teleplay expansion and subsequent graph edits

Safety and compatibility

• preserve episode, scene, and transition membership during remediation
• reject stale writes, invented identifiers, malformed sparse patches, new graph/playthrough/continuity blockers, and incomplete context reviews
• disable fallback after a prompt is budgeted for the selected provider so a smaller model cannot receive an oversized partial review
• migrate customized episode-weave prompts safely while preserving user overrides

Validation

• 250 focused server contract tests passed
• 36 FableLoom route tests passed
• 28 focused client interaction/API tests passed
• full client Biome lint passed across 2,375 files
• production client build passed
• prompt migration and data-drift checks passed
• three independent whole-diff review batches reported clean