Vouch request: generate_image per-request provider + Codex-subscription images

[insodimension]

Hi! Asking for a vouch to open a PR (per CONTRIBUTING.md — I had one merged PR before the vouch seed, so I'm not on the list yet: #2260, "Show active OAuth account in usage").

What I'd like to contribute (branch ready, written against current main):

1. Codex (ChatGPT subscription) images independent of the active chat model. Today generate_image only rides a Codex subscription when the active model is an OpenAI hosted-image model; otherwise it falls back to the metered OPENAI_API_KEY. This resolves a connected openai-codex OAuth credential directly (chatgpt.com/backend-api/codex/responses, chatgpt-account-id header), so images run on the subscription even when chatting with Claude/Gemini. providers.image gains an openai-codex option; auto detects a connected subscription (priority: active GPT image tool > Codex subscription > Antigravity > xAI > OpenRouter > Gemini).
2. Per-request provider param on generate_image (auto | openai | openai-codex | antigravity | xai | gemini | openrouter) — "generate this using gemini/xai" routes a single call without touching the global setting; absent → the setting applies, unchanged semantics.

Diff: main...insodimension:oh-my-pi-upstream-pr:feat/image-gen-provider (+312/−14 across tools/image-gen.ts, config/settings-schema.ts, tests, changelog). 7/7 image-gen.test.ts incl. 3 new tests (codex-subscription routing with a non-OpenAI active model, per-request override beating the setting, Responses edit path); check:ts green.

We run this in production in our downstream fork. Happy to adjust anything in review — e.g. the codex image-capable model priority ladder is a hardcoded const and could be made configurable if you prefer.

[can1357]

!vouch @insodimension
Tyvm!