fix(media): async submit (no more 524) + forward gpt-image size; remove dead KIE provider

[duanbing]

Why

Two production issues on media generation (surfaced by openai/gpt-image-2 text-to-image), plus a dead-code removal.

1. 524 on media generation

NovitaProvider::infer_media_proxy blocked the /inference HTTP response for the entire render, polling the upstream async task to completion before returning. Renders over ~100s tripped Cloudflare's origin timeout (524) even though the generation succeeded server-side and the user was charged.

Fix: return the task_id as soon as the task is submitted, and run poll_async_result + the result callback in a detached spawn_ignoring_shutdown task. This matches the callback-driven contract the backend already implements (submit → pending → callback). A new post_media_callback_failure reports async failures so a gen that errors after submit settles/refunds instead of hanging in pending.

2. gpt-image size dropped

The GptImageTextToImage / GptImageEdit request-shape whitelists omitted size, so the chosen dimension never reached Novita (and tier pricing fell back to the flat rate on the RouterBase side). Added size to both.

3. Remove unused KIE provider

No active model routes through KIE (type = "kie" is used by zero models; every media variant has a .proxy), so the None-proxy KIE media fallback was dead. Removes providers/kie.rs and the ProviderConfig::KIE / ProviderType::KIE / KIEKind / KIEProviderTypeConfig plumbing. Media variants without a proxy now return an explicit config error instead of silently falling back.

Also

• Added the missing media_generation: Default::default() to test InferenceParams initializers (the field was added earlier but ~28 test sites weren't updated), so the test target compiles again.

Verification

• cargo check --package tensorzero-core ✅
• cargo check --package tensorzero-core --tests ✅
• cargo fmt (unrelated workspace-wide reformatting reverted to keep the diff focused)

🤖 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.}