bug(kagent): Decommissioned Groq model llama3-8b-8192 mapped in setup.yaml

## Problem

`kagent-feast-mcp/manifests/kagent/setup.yaml` line 21 specifies:

```yaml
model:
```

This placeholder gives no guidance to new contributors. During local PoC testing, the previous value `llama3-8b-8192` returned a `400 model_decommissioned` error from the Groq API. A working default should be provided so the PoC is functional out of the box.

## Reproduction

```bash
# After applying setup.yaml with a decommissioned or empty model value:
kubectl logs -f deployment/kubeflow-docs-agent -n docs-agent
# → Groq API returns: 400 {"error": {"message": "model_decommissioned"}}
```

## Proposed Fix

Replace the placeholder with a known working model:

```yaml
# Before (line 21)
model:

# After
model: llama-3.1-8b-instant
```

Add a YAML comment listing other valid options for contributors who want to swap models.

## File

- `kagent-feast-mcp/manifests/kagent/setup.yaml:21`

## Context

Discovered during local Kagent PoC validation — [#59 (comment)](https://github.com/kubeflow/docs-agent/issues/59#issuecomment-3980450501)

Open kagent-feast-mcp/manifests/kagent/setup.yaml at line 21 and review the model setting and nearby manifest context. Apply the manifest in the local PoC and inspect the kubeflow-docs-agent logs with kubectl. Done means the setup uses a working default model, documents alternative valid options, and no longer produces the reported model_decommissioned error.

Written by the indexing model from the issue text.