feat(chatgpt_codex): add gpt-5.5 to known models

[michaelneale]

What

Adds gpt-5.5 to the CHATGPT_CODEX_KNOWN_MODELS list in the ChatGPT Codex (OAuth) provider so it shows up in the model picker, instead of users having to type it in by hand.

 pub const CHATGPT_CODEX_KNOWN_MODELS: &[ChatGptCodexModelAttrs] = &[
+    ChatGptCodexModelAttrs {
+        name: "gpt-5.5",
+        reasoning_levels: &["low", "medium", "high", "xhigh"],
+    },
     ChatGptCodexModelAttrs { name: "gpt-5.4", ... },
     ChatGptCodexModelAttrs { name: "gpt-5.3-codex", ... },
 ];

Placed first so it's the most prominent option in the picker.

Why

The ChatGPT Codex provider's available-models list is hand-curated (the OAuth backend at chatgpt.com/backend-api/codex doesn't expose a usable /models endpoint for this flow), so each new model needs an explicit entry. Users have been setting gpt-5.5 manually and reporting it works — this just surfaces it properly in the UI/CLI.

Verification

Tested live against the real ChatGPT Codex backend (POST https://chatgpt.com/backend-api/codex/responses) using the exact OAuth + payload shape create_codex_request builds:

Model	HTTP	completed.model	Tool calls
gpt-5.3-codex	200	gpt-5.4 (server alias)	✅
gpt-5.5	200	gpt-5.5	✅
gpt-5.4	200	gpt-5.4	✅

gpt-5.5 returns the same SSE event schema goose's stream parser already consumes (response.output_text.delta, response.function_call_arguments.done, etc.), so no other code paths needed changes. Reasoning effort low/medium/high/xhigh all accepted.

What was not changed

• CHATGPT_CODEX_DEFAULT_MODEL is still gpt-5.3-codex (it gets a special coding-agent preamble injected). Happy to flip it to gpt-5.5 in a follow-up if desired.
• No preamble special-case for gpt-5.5 — it gets the plain system prompt, same as gpt-5.4.

Tests

• cargo test -p goose --lib providers::chatgpt_codex — all 15 tests pass
• cargo fmt ✅
• cargo build -p goose ✅