Default to NRP deepseek-v4-flash, matching wyoming (#48)

Moves this app's default LLM off `qwen` (DSE-nimbus) onto `deepseek-v4-flash`,
the NRP-hosted model wyoming has been running.

Two edits, because the default alone was not enough: the picker carried only
the OpenRouter `deepseek/deepseek-v4-flash`, a different model on a different
provider, so `deepseek-v4-flash` is added as an explicit NRP option. It goes
first in `llm_models` so the `llm_models[0].value` fallback in agent.js
resolves to the same model as `llm_model` rather than disagreeing with it.

`qwen` stays in the picker as a selectable option.

Routing needs no proxy change: open-llm-proxy declares `deepseek-v4-flash`
explicitly under the NRP provider (exact match, with a `thinking` dialect
entry), so this is a config-only change.

Coordinated in geo-agent-ops (MIGRATIONS.md).

Co-authored-by: Default User <email@example.com>

Author: cboettig

k8s/configmap.yaml

@@ -14,8 +14,14 @@ data:
     {
         "mcp_server_url": "${MCP_SERVER_URL}",
         "mcp_auth_token": "${MCP_AUTH_TOKEN}",
-        "llm_model": "qwen",
+        "llm_model": "deepseek-v4-flash",
         "llm_models": [
+            {
+                "value": "deepseek-v4-flash",
+                "label": "NRP DeepSeek V4 Flash",
+                "endpoint": "/api/llm",
+                "api_key": "unused"
+            },
             {
                 "value": "qwen",
                 "label": "DSE-nimbus",