Add OpenAI-compatible video APIs and a SiftQ provider

Routine checks

• I have confirmed that there is no existing issue covering the same request.
• I have confirmed that I am using the latest version.
• I have read the project README and confirmed that the current version does not support this functionality.
• I am willing to follow up on this issue, assist with testing, and provide feedback.
• I understand that the maintainers have limited time and that issues that do not follow the project rules may be ignored or closed.

Feature description

Add two related modules to One API:

1. OpenAI-compatible asynchronous video APIs.
2. An independent SiftQ provider for the MiniMax-H3 video model.

The proposed public video API includes:

• POST /v1/videos to create a video generation job.
• GET /v1/videos/{video_id} to retrieve a job.
• GET /v1/videos to list the current user's jobs.
• GET /v1/videos/{video_id}/content to download completed video content.
• DELETE /v1/videos/{video_id} to cancel or delete a job.

The create endpoint should accept JSON and multipart/form-data requests with OpenAI-style fields such as model, prompt, seconds, size, and input_reference.

The SiftQ provider should include:

• Provider name: SiftQ.
• Model: MiniMax-H3.
• Default base URL: https://siftq.com/api/minimax/.
• Bearer API key authentication.
• A configurable base URL.
• MiniMax-H3 model-list registration.
• Channel configuration for the default, berry, and air themes.
• User-, token-, and channel-scoped asynchronous task ownership.
• Per-second video quota accounting with refunds when job creation fails.
• Normalized queued, in-progress, completed, and failed task states.
• Video content download and structured error handling.

Use cases

1. OpenAI-compatible clients can use a single One API token and base URL for asynchronous video generation.
2. Administrators can configure SiftQ video channels using the existing channel management, model permission, group, load-balancing, and quota features.
3. Persisted user and channel task ownership ensures that polling, listing, downloading, and deleting a video job use the channel that originally created it and cannot access another user's jobs.

An implementation branch is available for testing and review:

https://github.com/siftq/one-api/tree/feat/siftq-provider