feat(rubyllm): generate images via Context#paint behind use_ruby_llm flag

[fluxgravity]

RubyLLM image generation behind the use_ruby_llm flag

Completes the optional Phase 6 item skipped in #775: AIBackend::RubyLLM.generate_image wraps a keyed RubyLLM.context.paint instead of inheriting the #780 delegation to AIBackend::OpenAI.generate_image. Same { b64_json:, model: } contract and dispatch seam — flag off keeps the ruby-openai path byte-identical.

Related PRs

• RubyLLM phased migration feature-flagged #775 — completes its optional Phase 6 item
• Provider identity dispatch #797 (provider-identity-dispatch) — both directions:
  • Provider identity dispatch #797 scopes the old path to the canonical OpenAI URL; this branch keeps the name-based lookup, so reconcile both to URL-based selection when Provider identity dispatch #797 merges
  • Until Provider identity dispatch #797 lands, its stored per-backend choices are inert at dispatch, so image-gen routing here follows only the use_ruby_llm flag.

Known trade-off (deferred to #797)

A renamed OpenAI service keeps image gen flag-off but gets the "OpenAI API key not found" error flag-on. The URL-based reconciliation above resolves it.

Testing

• bin/rails test: 912 runs, 0 failures; old-path tests untouched.
• Live dogfood: image generated via a Claude chat; images/generations POST confirmed.
• Phase 7 (if implemented) can remove ruby-openai.

[fluxgravity]

@strivedi183 do you have any feedback on this? I deferred to your #797 for the URL-based selection refactor. If that gets merged first, then this branch will likely need a minor rebase tweak.