pin transformers in esm2 golden values (#1100)

ESM-2 model tests are currently failing on nightly since a transformers
commit broke modelopt from nemo. We really need to decouple our golden
value conversions from nemo.io, i suspect this will be an issue going
forward with all the stuff they import

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

## Summary by CodeRabbit

- Bug Fixes
- Fixed a compatibility issue that could cause installation or runtime
errors with the esm2 model by constraining a third-party dependency
version, improving reliability across environments.

- Chores
- Updated dependency constraints to ensure stable builds and predictable
behavior.
- No changes to visible features or workflows; users may need to
reinstall or update dependencies to pick up the fix.

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

Signed-off-by: Peter St. John <pstjohn@nvidia.com>

Author: pstjohn

models/esm2/pyproject.toml

@@ -19,7 +19,7 @@ dependencies = [
     "pytest",
     "torch",
     # "transformer_engine[pytorch]",
-    "transformers",
+    "transformers<4.56", # TODO: fix me, currently failing with a modelopt import from nemo.
 ]