Default to GPT-6 Luna and require live plugin acceptance

[jsoneaday]

Summary

• Add GPT-6 Luna/Sol/Astra, default profiles and helpers to gpt-6-luna, and constrain request fields (temperature/thinking) to what each model accepts.
• Allow asking the user questions during plugin create; factory reviews code before live hops and rejects fixture-only tests via type-based acceptance directions.
• Harden factory model streams so stalled builder replies time out instead of hanging the create dialog; route provider credit/limit failures through the create-plugin UI when present.

Test plan

• Confirm profile defaults and model picker show gpt-6-luna (and Sol/Astra) with valid thinking options
• Create a messaging plugin and verify the factory asks clarifying questions when needed, then runs live hops (not fixture http_results)
• Confirm a stalled factory stream surfaces a timeout instead of freezing on the last progress line
• Trigger a provider limit/credit failure and confirm it routes to the create-plugin UI path
• Run PluginLiveAcceptanceTests, FactoryModelStreamTimeoutTests, and related Structure/Plugin/LLMAgentClient unit tests

Made with Cursor