Upgrade OpenAI image generation to GPT Image

[Dereje1]

Motivation

• Replace the hard-coded DALL·E 3 image path with the GPT Image approach used elsewhere so generated images work with newer OpenAI responses while preserving the existing Pin creation and S3 upload flow.
• Keep the change minimal and backwards-compatible with the current frontend imgURL contract and existing processImage()/S3 logic.

Description

• Updated openai dependency range to ^4.52.1 in package.json and package-lock.json to match a GPT Image-capable SDK.
• Switched server image generation in server/controllers/post.ts to use OPENAI_IMAGE_MODEL = process.env.OPENAI_IMAGE_MODEL || 'gpt-image-1', replaced the hard-coded dall-e-3, and kept n: 1 and size: '1024x1024'.
• Added support for GPT Image b64_json responses by converting to data:image/png;base64,<base64> and retaining url as a fallback, and changed persisted AI-generated data to only store lightweight metadata (created, hasUrl, hasB64Json, model) rather than the full base64.
• Added/updated tests in tests/server/controllers/post.test.ts to cover the configured model call, base64 (b64_json) handling, URL fallback, missing-image payload, and verification that the full base64 payload is not persisted, and documented OPENAI_IMAGE_MODEL=gpt-image-1 in README.md.
• Files changed: package.json, package-lock.json, server/controllers/post.ts, tests/server/controllers/post.test.ts, README.md.
• Closes Upgrade OpenAI image generation from DALL-E 3 to GPT Image #66

Testing

• Ran the focused controller tests with npx jest tests/server/controllers/post.test.ts --runInBand and they passed.
• Ran the full test suite with npm test -- --watch=false --runInBand and all test suites passed.
• Ran TypeScript checks with npm run compileTS which succeeded, and lint with npm run lint which succeeded after a minor test-line-length fix.
• Built server and client with npm run build_server and npm run build_client, both completed successfully.
• Note: npm install openai@^4.52.1 --package-lock-only could not be completed in this environment due to an HTTP 403 from the registry/proxy, so the lockfile entries were updated in-place and all local automated checks used the updated files and mocked OpenAI behavior.

---

Codex Task