fix: remove unused DPO parameters from schema and tests (#2988)

# What does this PR do?

I removed these DPO parameters from the schema in [this
PR](https://github.com/meta-llama/llama-stack/pull/2804), but I may not
have done it correctly, since they were reintroduced in [this
commit](cb7354a9ce (diff-4e9a8cb358213d6118c4b6ec2a76d0367af06441bf0717e13a775ade75e2061dR15081))—likely
due to a pre-commit hook.

I've made the changes again, and the pre-commit hook automatically
updated the spec sheet.

tests/integration/post_training/test_post_training.py

@@ -195,10 +195,6 @@ class TestPostTraining:
         algorithm_config = DPOAlignmentConfig(
             beta=0.1,
             loss_type=DPOLossType.sigmoid,  # Default loss type
-            reward_scale=1.0,  # Scaling factor for reward signal (neutral scaling)
-            reward_clip=5.0,  # Maximum absolute value for reward clipping (prevents extreme values)
-            epsilon=1e-8,  # Small value for numerical stability
-            gamma=1.0,
         )
         data_config = DataConfig(
             dataset_id=dataset.identifier,