Add KIE.AI media provider support

[wauputr4]

Closes #1529

Summary

• Add KIE.AI as a provider across media capabilities, including image, video, music, text-to-speech, and transcription flows.
• Add a KIE task executor plus shared task utilities for callback URL defaults, task state normalization, result JSON parsing, and async polling.
• Wire KIE models into the dashboard media playground, provider registry, icon mapping, validation flow, and media handler tests.

Notes

• KIE market models keep their full provider model IDs when sent to the unified jobs API.
• Direct KIE music generation uses the direct generate/record-info polling endpoint while market jobs use jobs/recordInfo.
• Text-to-speech normalizes OpenAI-style voice aliases to KIE/ElevenLabs voice names and falls back to a callback URL when none is supplied.

Validation

• Ran formatter on changed KIE handler/config files.
• Pre-commit lint and explicit-any budget checks passed on the KIE review-fix commits.
• Manually exercised the media playground flow locally for the reported KIE media scenarios.

[gemini-code-assist]

Warning

Gemini is experiencing higher than usual traffic and was unable to create the review. Please try again in a few hours by commenting /gemini review.

[chatgpt-codex-connector]

💡 Codex Review

OmniRoute/open-sse/executors/index.ts

Line 56 in a591b4f

kie: new KieExecutor(),

Route KIE chat traffic through DefaultExecutor

Registering kie as a specialized executor here makes all chat requests for provider kie use KieExecutor instead of DefaultExecutor; KieExecutor is initialized with https://api.kie.ai and does not override chat URL construction, so requests are sent to the API root rather than the configured chat-completions URL, which breaks KIE chat/completions calls in production for this provider.

---

OmniRoute/open-sse/handlers/imageGeneration.ts

Lines 467 to 470 in a591b4f

	payload = {
	prompt,
	image_size: mapImageSize(size, "1:1"),
	num_images: body.n || 1,

Send direct KIE image params with correct field names

For non-market KIE image models (for example kie/gpt4o-image), this payload uses image_size and num_images, but the direct KIE generate flow in this change expects OpenAI-style size and nVariants; as a result direct image requests can ignore requested size/count or fail upstream validation, even though the market-model path works.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[wauputr4]

Addressed the Codex review feedback in commit 667ce4db:

• Removed kie from the specialized chat executor map so KIE chat/completions traffic routes through DefaultExecutor and the configured provider registry URL.
• Kept KieExecutor available for media handlers that need task create/poll behavior.
• Updated direct KIE image payloads to send size and nVariants, matching the direct KIE image flow and existing handler test expectations.
• Added a regression test to guard that KIE chat routing continues to use the default executor while media can still import the KIE task executor.

Pre-commit checks passed locally: prettier, eslint, docs-sync, and explicit-any budget.