Current GPT Model Budgets (openai + azure_openai) #538

As a SkillSpector user on OpenAI or Azure OpenAI, I want the bundled registries to know current GPT models — including GPT-6 Astra — so semantic scans of complex code get accurate token budgets instead of the 128K fallback.

The gap

The openai registry knows only gpt-5.4 (at a stale 1M context) and the azure_openai registry stops at GPT-4, so current models silently fall back to the package-wide default budget. Azure users on private endpoints are affected the same way, since deployments resolve by model name.

What I am asking for

Please consider adding the verified current generation to both bundled registries: the GPT-5 line at 400K/128K and the 1.05M line (gpt-5.4 corrected, gpt-5.5, the 5.6 family, gpt-6-Astra) at 1.05M/128K, with lookup tests in the existing style. Done when get_context_length/get_max_output_tokens return documented budgets for these IDs on both providers, with no code or default-model changes.

This could be achieved by

• Adding the 13 entries per provider to src/skillspector/providers/openai/model_registry.yaml and src/skillspector/providers/azure_openai/model_registry.yaml, each block citing its official source and verification date
• Annotating the -codex entries as Responses-API-only (this provider speaks chat completions) and omitting IDs with no general release (base gpt-5.3) or unverified numbers (gpt-4.1, o-series)
• Extending tests/unit/test_providers.py and tests/unit/test_new_providers.py with test_metadata_gpt5_generation / test_metadata_flagship_generation loops