Switch public inference default model to DeepSeek in Agentic Env Gen (#1181)

## Summary
Switch public inference model to DeepSeek

## Detailed description
- gpt-oss-120b has been deprecated today without prior notice, caught by
SQA regression tests
- Switched to deepseek-v4-pro as it has been tested on all 5 example
workflows
- The inference call time increases to 100+sec from 10+sec, but given
the quality outperforms than nemotran, the cost is acceptable.
- Updated doc to track perf

---------

Signed-off-by: Xinjie Yao <xyao@nvidia.com>

Author: xyao-nv

docs/pages/concepts/agentic_environment_generation/model_selection.rst

@@ -60,10 +60,10 @@ resolution.
      - Mean runtime
    * - ``public`` (default)
      - Public (free)
-     - ``openai/gpt-oss-120b``
+     - ``deepseek-ai/deepseek-v4-pro-0813``
      - ``NVIDIA_API_KEY``
-     - 13/15 (86.7%)
-     - 22.28 s
+     - 15/15 (100%)
+     - 150.66 s
    * - ``internal``
      - NVIDIA internal
      - ``openai/openai/gpt-5.6-terra``
@@ -81,7 +81,10 @@ resolution.
    The benchmark ran each of five documented prompts three times. Pass rate is the fraction of
    generated specs that matched the expected structure; runtime is the mean end-to-end
    ``generate_spec`` runtime. These results are snapshots rather than guarantees: model output is
-   non-deterministic, and service load affects runtime.
+   non-deterministic, and service load affects runtime. On the public endpoint, overriding the
+   default with ``--model nvidia/nemotron-3-ultra-550b-a55b`` passed 9/15 runs (60%) with a
+   149.86-second mean runtime; ``--model openai/gpt-oss-20b`` passed 7/15 runs (46.7%) with a
+   333.17-second mean runtime.
 
 Reviewing the Generated Spec
 ----------------------------

isaaclab_arena/agentic_environment_generation/inference_backend.py

@@ -50,8 +50,7 @@ class InferenceEndpoint:
 PUBLIC_ENDPOINT = InferenceEndpoint(
     name="public",
     base_url="https://integrate.api.nvidia.com/v1",
-    # If you cannot access the model in your region, try "nvidia/nemotron-3-super-120b-a12b".
-    model="openai/gpt-oss-120b",
+    model="deepseek-ai/deepseek-v4-pro-0813",
     api_key_env_var="NVIDIA_API_KEY",
 )
 """Publicly reachable build.nvidia.com endpoint, reached with an NGC API key."""