fix lm-eval version conflict in puzzletron requirements (#1257)

## Summary

- lm-eval 0.4.10 has API changes that break `lm_eval_hf.py`, which
monkey-patches `HFLM.create_from_arg_obj`
- Pin `examples/puzzletron/requirements.txt` from `lm-eval==0.4.10` to
`lm-eval==0.4.8` to match `examples/llm_eval/requirements.txt`
- Add version compatibility warning to `lm_eval_hf.py` for early
detection if versions drift again



<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

* **Chores**
* Updated the lm-eval dependency version to 0.4.8 in the Puzzletron
example
* Added compatibility version checking to the LM Eval example with
appropriate notifications

<!-- end of auto-generated comment: release notes by coderabbit.ai -->

Signed-off-by: jrausch <jrausch@nvidia.com>
Signed-off-by: Keval Morabia <28916987+kevalmorabia97@users.noreply.github.com>
Co-authored-by: Keval Morabia <28916987+kevalmorabia97@users.noreply.github.com>

Author: j-rausch

examples/llm_eval/lm_eval_hf.py

@@ -40,8 +40,15 @@
 import warnings
 
 import datasets
+import lm_eval
 from lm_eval import utils
 from lm_eval.__main__ import cli_evaluate, parse_eval_args, setup_parser
+
+if not lm_eval.__version__.startswith("0.4.8"):
+    warnings.warn(
+        f"lm_eval_hf.py is tested with lm-eval 0.4.8; found {lm_eval.__version__}. "
+        "Later versions may have incompatible API changes."
+    )
 from lm_eval.api.model import T
 from lm_eval.models.huggingface import HFLM
 from quantization_utils import quantize_model

examples/puzzletron/requirements.txt

@@ -1,4 +1,4 @@
-lm-eval==0.4.10
+lm-eval==0.4.8
 math-verify
 ray
 # Likely works for transformers v5 also, but we need to test it