Support /fast with OpenAI API-key authentication

[joshkmartinez]

/fast currently only works with ChatGPT authentication, but not with OpenAI API-key authentication.

The OpenAI Responses provider already forwards service_tier and accounts for priority-processing costs. Could /fast also enable service_tier: "priority" for eligible OpenAI API models?

I have an implementation that:

• allows eligible openai / openai-responses models to use /fast
• preserves the existing session and child-agent behavior
• sends "service_tier": "priority" in the OpenAI request
• keeps the existing priority-processing cost calculation

its on my fork. I’m happy to wait for maintainer direction before submitting it.

[snimu]

Done in #1609