Cannot use Grok 4.20 due to reasoningEffort being sent always for responses wire format

### Describe the bug

I'm setting the following envvars:

```
Set COPILOT_PROVIDER_TYPE=openai
Set COPILOT_PROVIDER_BASE_URL=https://api.x.ai/v1
Set COPILOT_MODEL=grok-4.20-non-reasoning
Set COPILOT_PROVIDER_WIRE_API=responses
Set COPILOT_PROVIDER_MAX_PROMPT_TOKENS=2000000
Set COPILOT_PROVIDER_API_KEY=***********
```

There seems to be no way to turn that parameter off entirely (or even customize as proposed at https://github.com/github/copilot-cli/issues/2559).

Ideally there should be support for setting `off` (don't send the parameter at all) as well as custom values via `COPILOT_MODEL_EFFORT` envvar

NOTE: despite 4.20 and 4.1 being reasoning models, the parameter is still not supported. See https://docs.x.ai/developers/model-capabilities/text/reasoning#the-reasoning-parameter

### Affected version

1.0.34

### Steps to reproduce the behavior

Set the above variables so they are valid for xAI (in particular your API key).

### Expected behavior

Requests succeed.

### Additional context

_No response_

Empieza rastreando cómo COPILOT_PROVIDER_WIRE_API=responses construye las solicitudes y dónde se establece reasoningEffort a partir de la configuración del modelo. Comprueba el manejo del proveedor y del esfuerzo del modelo con respecto al comportamiento de xAI descrito en el issue; se considera terminado cuando se admite una configuración off o un valor COPILOT_MODEL_EFFORT personalizado, de modo que la solicitud se complete correctamente sin un parámetro no compatible.

Escrito por el modelo de indexación a partir del texto del issue.