fix: setup wizard overwrites custom endpoint config with OpenRouter defaults

[patlegu]

Problem

When selecting Custom OpenAI-compatible endpoint in the setup wizard, the settings entered by the user (base URL, model name) are silently overwritten by the OpenRouter defaults after the wizard completes.

Symptoms: ~/.hermes/config.yaml ends up with:

model:
  base_url: https://openrouter.ai/api/v1
  default: anthropic/claude-opus-4.6

instead of the user-provided values (e.g. http://localhost:11434/v1 + local model).

Root Cause

_model_flow_custom() (main.py) loads its own independent copy of config via load_config(), applies the custom endpoint settings, and saves it. However, the caller setup_model_provider() (setup.py) holds a separate config dict loaded at wizard startup. The final save_config(config) call at line 1802 of setup.py overwrites everything _model_flow_custom() just saved.

Fix

Reload config in setup.py immediately after _model_flow_custom() returns, so the caller's copy reflects the saved changes before the final write.

from hermes_cli.main import _model_flow_custom
_model_flow_custom(config)
# _model_flow_custom saves config independently (loads its own copy).
# Reload here so the final save_config(config) below doesn't overwrite it.
from hermes_cli.config import load_config as _reload_config
config.update(_reload_config())

One line added, no refactoring. _model_flow_custom() logic is untouched.

Reproduction

1. Run hermes setup
2. Select Custom OpenAI-compatible endpoint
3. Enter http://localhost:11434/v1 as base URL, ollama as API key, qwen2.5-coder:14b as model
4. Complete the wizard
5. cat ~/.hermes/config.yaml → model.base_url is still https://openrouter.ai/api/v1

Tested with Ollama on WSL2.