Honor provider config and support GPT-5.6 models

[RyanKung]

Summary

• Make serve honor the persisted config.provider before falling back to all saved providers.
• Keep explicit CLI --provider as the highest-priority selection.
• Add gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna to the Codex model registry and highlight ordering.
• Bump rotom to 1.5.7 and update README examples to use gpt-5.6-sol.
• Normalize legacy minimal reasoning effort to low for GPT-5.6 Codex requests.
• Fix the Cursor built-in decline path for the latest stable clippy question_mark lint.

Root Cause

resolve_served_providers only considered the persisted runtime provider when no credentials were saved. Once multiple provider credentials existed, serve built upstreams for every saved provider, so stale unrelated credentials such as Cursor could still be loaded/refreshed even when the user had configured a specific provider.

For GPT-5.6, OpenAI API documentation lists the gpt-5.6 alias, but the Codex/ChatGPT OAuth backend currently rejects that alias for this account while accepting the concrete gpt-5.6-* model IDs. The built-in list therefore exposes only the concrete IDs verified against the Codex backend.

CI also caught a latest-stable clippy warning in existing Cursor code: the final optional built-in fetch branch can use the ? operator instead of an else { return None; } tail.

Impact

A configured provider now scopes the served upstreams and token status refresh tasks, avoiding background refresh attempts for unrelated stale providers. Codex users can select the GPT-5.6 Sol, Terra, and Luna model IDs directly through /v1/models, rotom models, and Anthropic/OpenAI-compatible clients. The default Anthropic fallback remains gpt-5.5.

Fixes #2.

Validation

• cargo fmt --all
• cargo clippy --all-targets --locked -- -D warnings
• cargo test --all-targets --locked
• cargo run -- --version -> rotom 1.5.7
• cargo run -- models --provider openai
• cargo run -- status --provider openai
• Local gateway chat completion smoke tests returned OK for gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna
• Confirmed gpt-5.6 alias is rejected by the Codex OAuth backend and is not exposed in the built-in model list
• GitHub Actions CI #45 passed on PR head 49cea9c