img-forge: propagate OpenAI direct-model contract and ultra-tier preflight

[stackbilt-admin]

Context

img-forge now treats openai_model as direct provider selection, not a tier/quota bypass. Supported values are gpt-image-1 and dall-e-3; both must be surfaced to users as OpenAI execution gated/metered as ultra tier.

Required external work

• Update any proxied image_generate / img-forge tool schema to accept openai_model values: gpt-image-1, dall-e-3.
• Ensure the proxy forwards openai_model unchanged to img-forge.
• Ensure delegated calls still include X-Tenant-User-Id so img-forge can preflight quota as ultra before provider spend.
• Update tool descriptions/docs to avoid saying OpenAI bypasses quality tiers; it routes directly to OpenAI but uses ultra-tier quota and billing.
• Add/adjust a regression test that a delegated OpenAI request is represented as an ultra-tier request or otherwise reaches img-forge with the tenant context intact.

Acceptance

A consumer can request openai_model: "gpt-image-1" or openai_model: "dall-e-3" through the MCP gateway, and the request is forwarded with tenant delegation intact so img-forge performs the ultra-tier quota gate before using the admin provider key.