Add machine-local character voice profiles using Kokoro and Piper

[atomantic]

Part of #5377. Blocked by #5378.

Goal

Add machine-local, approved character voice profiles that initially wrap the existing Kokoro and Piper voices and provide a stable resolver for offline dialogue and hosted-mode prototypes.

Proposed approach

• Add DB-primary voice_profiles metadata with managed files under data/voice-profiles/<profileId>/, documented in docs/STORAGE.md and included in backup.
• Bind profiles locally by (universeId, characterId) and route (studio or interactive) without persisting profile ids into federated Universe records.
• Extend the TTS facade with capability discovery and profile-aware synthesis while keeping preset synthesis backward compatible.
• Add benchmark lines, preview comparison, explicit promotion, revisions, and approval state.
• Add a reproducible mastering chain and capability-probed rate, modest pitch, and formant controls; disable unsupported controls with setup guidance.
• Resolve approved profile -> existing namespaced voiceId -> project/system fallback, reporting degraded bindings.

Acceptance criteria

• Existing Kokoro/Piper synthesis continues to work without profile migration.
• A character can promote an existing preset as an approved profile and render the fixed benchmark script.
• Offline dialogue provenance records profile revision, engine/model revision, effective delivery controls, timings, and mastering.
• Changing the project default voice does not change a character with an approved profile.
• Source assets and artifacts remain machine-local, use safe basenames, and are backed up.
• Missing, unavailable, and intentionally empty states remain distinct throughout resolver and UI behavior.

Out of scope

• New voice-model downloads, cloning, or training.
• Automatic approval based on similarity metrics.