Add GPT-5.6 family support

[michaelneale]

Summary

• register gpt-5.6-luna, gpt-5.6-sol, and gpt-5.6-terra with their 1,050,000-token context windows
• add the missing GPT-5.5 and GPT-5.5 Pro static context metadata
• recognize GPT-5.6 models as supporting xhigh reasoning effort
• add focused tests for model context metadata and reasoning effort handling

Verification

• Probed the OpenAI API: all three GPT-5.6 models report a 922,000-token input limit, consistent with 1,050,000 total context minus 128,000 max output tokens
• Probed reasoning.effort: xhigh: all three GPT-5.6 models accepted it
• cargo test -p goose-provider-types test_create_request_gpt56_max_effort_uses_xhigh
• cargo test -p goose-providers gpt_5_5_and_5_6_models_have_expected_context_limits
• cargo clippy -p goose-provider-types -p goose-providers --all-targets -- -D warnings
• cargo fmt --all

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 54793f96d9

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4590c32718

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Add canonical records before listing GPT-5.6

These IDs are only added to OPEN_AI_KNOWN_MODELS, but the refreshed inventory and session materialization paths rely on the canonical registry: fetch_recommended_models filters live /v1/models results through canonical mappings when any mapped OpenAI models exist, and with_canonical_limits uses the same registry for the session context limit. I checked rg "gpt-5\\.6|gpt-5-6" crates/goose-provider-types/src/canonical/data and there are no canonical records, so once OpenAI returns these IDs they will be dropped from refreshed recommended models; if entered manually/fallback, Goose falls back to 128k instead of the 1,050,000 limit listed here. Add matching canonical entries with the new models' limits/capabilities when exposing them.

Useful? React with 👍 / 👎.

[alexhancock]

Nice @michaelneale