feat(novita): add ByteDance Seedream text-to-image request shapes

[duanbing]

Adds three NovitaRequestShape variants for the Seedream family's synchronous image endpoints. RouterBase seeds four models across them (main repo PR bumps the submodule pointer to this branch).

Variant	Endpoint	Allowed params
SeedreamV3TextToImage	/v3/seedream-3-0-txt2img	size, seed, guidance_scale, response_format, watermark
SeedreamV4TextToImage	/v3/seedream-4.0	size, sequential_image_generation, max_images, watermark
SeedreamImage (4.5 + 5.0-lite)	/v3/seedream-4.5, /v3/seedream-5.0-lite	size, sequential_image_generation, watermark

Notes:

• Synchronous endpoints (no async task). parse_urls already handles their image_urls / images response keys, so no response-side change.
• Seedream 3.0 additionally requires a fixed body model = "seedream-3-0-t2i-250415", which is forced server-side so a caller can't select a different checkpoint via their own model.
• 4.5 and 5.0-lite share one shape (identical param surface); the model is selected by URL slug carried in source_model_id.

Unit tests cover the 3.0 model-injection + allow-list filtering, the shared 4.5/5.0-lite shape, and the prompt-required guard. cargo check / clippy / the new tests pass.

🤖 Generated with Claude Code

[github-actions]

Thank you for your submission, we really appreciate it. Like many open-source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution. You can sign the CLA by just posting a Pull Request Comment same as the below format.

---

I have read the Contributor License Agreement (CLA) and hereby sign the CLA.

---

_{You can retrigger this bot by commenting recheck in this Pull Request. Posted by the CLA Assistant Lite bot.}