[M5] Integrate selected AI provider into model_generating pipeline

[weiuou]

Goal

Use the selected AI provider output in the real worker pipeline while preserving procedural fallback and the downstream paperability/export flow.

Scope

• Replace direct deterministic mesh generation inside model_generating with provider orchestration.
• Accept successful AI provider output as base_mesh and preview_model artifacts.
• Run minimal normalization: center, scale, reject empty meshes, reject no-face meshes, and record normalization status.
• Preserve procedural fallback for unconfigured provider, provider failure, invalid mesh, or normalization failure.
• Continue using existing downstream repair, decimation, unfolding, export, metrics, and next-action behavior.

Acceptance Criteria

• Worker tests cover AI success, AI failure fallback, and bad mesh fallback.
• Task status returns AI provider metadata on generated artifacts.
• Current full pipeline tests still pass with AI disabled.