No llm models were loadederror when not specifying `--model` parameter

# No llm models were loaded error when not specifying `--model` parameter

## Environment
- **aks-agent version:** 1.0.0b11
- **Azure CLI version:** 2.81.0
- **OS:** Linux

## Description
When running `az aks agent` without the `--model` parameter, the agent fails with "No llm models were loaded" even though the config file (`~/.azure/aksAgent.yaml`) is valid and contains correct Azure OpenAI credentials.

## Steps to Reproduce

1. Configure LLM via `az aks agent-init` or manually create `~/.azure/aksAgent.yaml`:

```yaml
llms:
- provider: azure
MODEL_NAME: gpt-4.1
AZURE_API_KEY:
AZURE_API_BASE: https://.openai.azure.com/
AZURE_API_VERSION: 2025-01-01-preview
```

2. Run without `--model`:

```bash
az aks agent "test" --no-interactive
```

3. Observe error:

```
No llm models were loaded
ERROR: Failed to create AI executor: No llm models were loaded
```

## Expected Behavior
The agent should load the model from the config file automatically.

## Actual Behavior
The model is not loaded. Debug output shows:

```
INFO: cli.azext_aks_agent.agent.llm_config_manager: Using provider: azure, model: None, Env vars setup successfully.
```

## Suspected Root Cause

In `azext_aks_agent/custom.py`, the `export_model_config()` function returns the model string, but it's not captured:

```python
llm_config_manager.export_model_config(llm_config) # Returns model but not captured!

aks_agent_internal(
...
model, # This is still None from command line args
...
)
```

Should be:

```python
model = llm_config_manager.export_model_config(llm_config) or model
```

## Workaround

Explicitly pass the `--model` parameter:

```bash
az aks agent "test" --model azure/gpt-4.1
```

## Additional Context

- The Azure OpenAI deployment is correctly configured (verified via direct API call)
- The config file is correctly parsed (verified via Python yaml.safe_load)
- Environment variables are correctly set by `export_model_config()`
- The issue is specifically that the model name is not passed to the Holmes config