Make provider setup append models instead of overwriting

[luckeyfaraday]

Problem

codex-shim setup <provider> wrote a single-model payload and overwrote the provider settings file. Running setup again to configure a second model silently dropped the first, so only the most recently set-up model appeared in the picker.

This bit a real session: setting up xiaomi/mimo-v2.5 overwrote a previously configured z-ai/glm-5.2 in openrouter-models.json. Because the shim reloads that file per request, the still-running GLM Codex session immediately started failing with 404 Unknown model slug/model: z-ai-glm-5-2 from the same shim that had been serving it.

Change

setup now merges the entered model into the existing rows, keyed by model id:

• a new model id is appended — every configured model keeps appearing in the picker;
• an existing model id updates that row in place — still idempotent, Enter-to-keep behavior preserved.

Existing rows are read as raw JSON (not via ModelSettings) so unknown / camelCase fields (extra_headers, no_image_support, etc.) on other models survive the rewrite untouched.

Tests

• test_provider_setup_appends_new_model — two setups → both models present, in order.
• test_provider_setup_updates_existing_model_in_place — same id twice → one row, fields updated.
• test_provider_setup_preserves_unknown_fields_on_other_models — appending leaves another model's extra_headers/no_image_support intact.

Full suite: 45 passed (with pytest-asyncio installed).

Docs

README provider-workflow section updated to describe add-vs-update and how to remove a model.

🤖 Generated with Claude Code