feat(content-studio): select scene video mode and bridge OpenAI images

[thxforall]

Parent: #508
Depends on: #519

Context

The landed job pipeline can generate scene clips, but it does not yet decide the right generation mode per scene or connect previously generated OpenAI images into image-to-video/reference-to-video flows.

Scope

• Add scene-level generation-mode selection for text-to-video, image-to-video, and reference-to-video.
• Use ShortFormPlan scene metadata, post originals, and generated Content Studio image assets when deciding the mode.
• Connect OpenAI image outputs to the video pipeline as image-to-video inputs.
• Keep scene retry behavior working per generated mode.
• Surface enough metadata in job/scene records to explain why a mode was selected.

Acceptance Criteria

• Scenes with usable generated/reference images can run through image-to-video or reference-to-video instead of defaulting to text-to-video.
• Scenes without image inputs still fall back to text-to-video.
• Focused tests cover mode selection, fallback, and retry behavior.

Out of Scope

• Final multi-clip composition/export.
• provider implementation beyond the abstraction from feat(content-studio): add provider-agnostic video provider abstraction #519.
• governance and cost tracking.