fix(openai): support image edit accounting and task image responses

[Brisbanehuang]

Summary

• Count image_edit.completed SSE events as completed image outputs while continuing to ignore partial edit image events.
• Support OpenAI-compatible image providers that return asynchronous task responses from /v1/images/generations: detect task_id, poll /v1/tasks/{task_id}, download final image URLs, and convert them back to OpenAI Images-compatible b64_json responses.
• Add tests for both image_edit.completed accounting and task-based image generation results.

This helps with providers such as APIMart whose gpt-image-2 generation endpoint returns a task object first, with final images available through the task status endpoint. Docs: https://docs.apimart.ai/cn/api-reference/images/gpt-image-2/generation

Tests

• go test ./internal/service -run 'TestOpenAIImageOutputCounter|TestOpenAIGatewayServiceForwardImages_APIKeyGenerationPollsTaskResult|TestOpenAIGatewayServiceForwardImages_APIKeyGenerationUsesConfiguredV1BaseURL|TestOpenAIGatewayServiceForwardImages_APIKeyStreamJSONResponseBillsImage'\n- go test ./internal/service\n

[github-actions]

All contributors have signed the CLA. ✅
_{Posted by the CLA Assistant Lite bot.}

[Brisbanehuang]

I have read the CLA Document and I hereby sign the CLA

[Brisbanehuang]

recheck