[codex] support gpt-image-2 image generation

[hxdflying]

Summary

Adds OpenAI gpt-image-2 support to the image generation path and exposes the generation options needed to use it effectively.

• detect gpt-image-2 in the OpenAI image provider and pass custom widthxheight sizes directly
• add provider-level quality support and thread image_quality through settings, pipeline, and VisualizerAgent
• validate OUTPUT_RESOLUTION / IMAGE_QUALITY and load image.quality from YAML
• expand high-resolution aspect-ratio dimension handling for methodology diagrams
• add pricing entries for gpt-5.5 and a conservative gpt-image-2 per-image estimate for the existing cost tracker model
• omit temperature for GPT-5-family OpenAI VLM calls, which use the API default

Validation

• git diff --cached --check
• python3 -m pytest tests/test_config.py tests/test_agents/test_visualizer.py tests/test_features.py tests/test_cost_tracking.py - 94 passed
• python3 -m pytest - 647 passed, 4 skipped, 1 warning

Notes

gpt-image-2 pricing is token-based, but PaperBanana currently records image generation pricing as a single per-image float. The added 0.211 value is a high-quality square-image estimate so cost tracking no longer treats the model as unknown.