[RFC] providers: add Zhipu BigModel for text, image, video, and speech generation #234

Minimal support

A user with one Zhipu BigModel account and API key should be able to write a script, generate an image, create a video, and synthesize narration in one OpenDirector project. Provide at least one supported model per capability. Generated media can be saved, adopted, and edited in the project.

Sub-issues

• Chat · GLM streaming chat and tool calling.
• Image · Synchronous text-to-image generation with GLM-Image.
• Video · CogVideoX-3 only. Support text-to-video, single-image-to-video, and first-and-last-frame generation. Persist the remote task ID for recovery and register completed media as a candidate.
• TTS and voices · GLM-TTS with preset voices for the minimum workflow. Cover voice cloning, listing, selection, and deletion in the voice work.

Shared provider configuration

Reuse OpenDirector's connection and credential storage across the four model categories. Scope model edits and deletions to the selected category and keep credentials isolated across accounts. The canvas and Agent share model capabilities, validation, and generation workflows.

Acceptance criteria

• With a fresh configuration and one Zhipu BigModel account and key, generate text, an image, a CogVideoX-3 video, and narration with a preset voice. Verify media adoption and reuse after reopening the project.
• Verify the shared canvas and Agent validation and result handling.
• Verify account isolation and category-scoped configuration changes.
• Verify the setup and error messages in English and Chinese.

Zhipu BigModel image API · Zhipu BigModel video API

Pull request evidence

Include screenshots in the PR description showing the implemented workflow running in OpenDirector and its result so reviewers can see how it works.