QVAC-20424 feat[api]: img2vid for POST /v1/videos

[lauripiisang]

🎯 What problem does this PR solve?

• POST /v1/videos only supported text-to-video; callers had no way to animate a still image via the CLI OpenAI server.

📝 How does it solve it?

• Accepts multipart/form-data on POST /v1/videos alongside the existing JSON body (no breaking change — JSON txt2vid continues to work unchanged).
• Mode is inferred from the presence of an init_image file field: provided → img2vid, absent → txt2vid.
• strength (0–1) controls denoise intensity for img2vid; coerced from string for multipart compatibility.
• Invalid strength values return 400 invalid_strength.
• Updated packages/cli/docs/serve-openai.md and docs/website/content/docs/ai-capabilities/video-generation.mdx to document both modes and the I2V model family.

🧪 How was it tested?

• Unit tests cover schema acceptance/rejection of init_image and strength, extractVideoCreateParams mode selection, strength coercion and range validation (371/371 pass).
• TypeScript compilation verified with the dev SDK build from the PR QVAC-19845 feat[bc|api]: add img2vid (image-to-video) support to video generation in SDK #2436 merge run. The dev build was installed locally as a package alias (not committed) — npm install @qvac/sdk@npm:@tetherto/sdk-mono@0.12.2-tmp.runid-27142936775 — and npx tsc --noEmit passed clean. package.json was restored before committing.
• E2e / functional tests are not included — actual img2vid execution requires a model loaded with clipVisionModelSrc which needs hardware. These should be added as a follow-up.

🔌 API Changes

POST /v1/videos — new optional fields (multipart only for init_image):

// txt2vid (unchanged):
fetch('/v1/videos', {
  method: 'POST',
  headers: { 'Content-Type': 'application/json' },
  body: JSON.stringify({ model: 'wan', prompt: 'a cat surfing' })
})

// img2vid (new):
const form = new FormData()
form.append('model', 'wan')
form.append('prompt', 'the subject slowly turns and smiles')
form.append('init_image', imageBlob, 'frame.png')
form.append('strength', '0.85')
fetch('/v1/videos', { method: 'POST', body: form })

⚠️ Merge blocker

Do not merge until @qvac/sdk is published with img2vid support (from PR #2436). The CLI routes VideoClientParams at runtime through the SDK — without the released types and execution pipeline, img2vid requests will fail at the SDK call site. Once a public SDK release including those changes is available, update @qvac/sdk in packages/cli/package.json to that version and re-run bun run build before merging.