fix(setup): defer config.yaml write until after model selection

[ygd58]

Problem

Closes #1182

_update_config_for_provider() was called immediately after provider selection for zai, kimi-coding, minimax, minimax-cn, and anthropic — before model selection happened. Since the gateway re-reads config.yaml per-message, this created a race condition where the gateway would pick up the new provider but still use the old (incompatible) model name.

Example: switching from OpenRouter (anthropic/claude-opus-4.6) to MiniMax would briefly write provider: minimax + base_url: https://api.minimax.io/v1 to disk while model.default was still anthropic/claude-opus-4.6. Any gateway message sent during setup would fail.

Even without a race: selecting "Keep current" at model selection would leave an OpenRouter-formatted model name permanently on a non-OpenRouter provider.

Fix

Capture selected_base_url in each affected provider block, then call _update_config_for_provider() once, after model selection completes, just before save_config(). The in-memory _set_model_provider() calls are kept in place so the config object stays consistent throughout the setup flow.

Affected providers

• zai
• kimi-coding
• minimax
• minimax-cn
• anthropic

[ygd58]

Closing — upstream has landed a partial mitigation in main (adding default_model parameter to _update_config_for_provider). This addresses the "Keep current" edge case but the race condition window still exists since config.yaml is written before model selection completes.

The full Option A fix (deferring the write until after model selection) remains the correct solution if the race condition is a priority. Happy to reopen if needed.

[teknium1]

Cherry-picked into PR #1499 with authorship preserved. Clean apply onto current main — the race condition was still present. Thanks for the fix, @ygd58!