provider: add image generation support for OpenRouter

[Robinnnnn]

Description: (required)

• Adds imageGenerate support to the OpenRouter provider, wiring it to OpenRouter's dedicated image generation endpoint (POST https://openrouter.ai/api/v1/images), which is OpenAI-Images-compatible.
• src/providers/openrouter/api.ts: route imageGenerate to /v1/images (resolves to https://openrouter.ai/api/v1/images).
• src/providers/openrouter/imageGenerate.ts (new): request config mapping OpenAI Images params (prompt, model, n 1-10, size) plus OpenRouter-specific params (seed, aspect_ratio, resolution, input_references for image-to-image). response_format is intentionally not mapped — OpenRouter always returns base64 (b64_json) and ignores/strips unknown params. Response transform follows the OpenAI/Deepbricks pattern: pass through the already-OpenAI-shaped success body ({created, data: [{b64_json}], usage}), and normalize upstream errors via OpenAIErrorResponseTransform (OpenRouter errors are {error: {message, code}}).
• src/providers/openrouter/index.ts: register the config and response transform.

Tests Run/Test cases added: (required)

• Live end-to-end through a locally running gateway (tsx src/start-server.ts) with a real OpenRouter API key against google/gemini-2.5-flash-image:
  • POST /v1/images/generations with {model, prompt, n: 1} → 200, data[0].b64_json decodes to a valid PNG (checked magic bytes), usage passed through.
  • size: "1024x1024" passthrough → 200, valid PNG.
  • Error normalization: unknown model → 404 {"error":{"message":"openrouter error: No model found...","code":404},"provider":"openrouter"}; conflicting size + aspect_ratio → 400 with OpenRouter's message, both in the gateway's standard error shape.
• tsc --noEmit (no new errors; two pre-existing errors in bytez/open-ai-base untouched), npm run build succeeds, prettier --check clean, npm run test:gateway results identical to main (47 passed; the 8 pre-existing suite failures on main are unchanged).

Type of Change:

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Documentation update
• Refactoring (no functional changes)

[Robinnnnn]

@VisargD @narengogi @b4s36t4 would appreciate a review! 🙏

This adds imageGenerate support to the OpenRouter provider, routed to OpenRouter's dedicated OpenAI-Images-compatible endpoint (POST /api/v1/images). It follows the existing OpenAI/Deepbricks pattern: map the OpenAI Images params (plus OpenRouter extras seed/aspect_ratio/resolution/input_references) and normalize upstream errors via OpenAIErrorResponseTransform. Like every other provider here, model is a passthrough param, so this exposes all of OpenRouter's image models with no hardcoded list to maintain.

Verified locally: tsc --noEmit clean on the new files (the 2 pre-existing bytez/open-ai-base errors are untouched), npm run build succeeds, and prettier --check passes on all 3 changed files (the CI formatting check should go green once the fork-PR workflow is approved to run). Happy to adjust anything.