Adjust reward model's score module and pooler module order for reducing computation

[aqweteddy]

Motivation

• @merrymercy mentioned in Gemma2 reward model support #1954
• remove redundant code in gemma2 reward model.

Modifications

• Adjust the order of the pooler and score modules in class LlamaForSequenceClassification and class Gemma2ForSequenceClassification to reduce computation.
• Remove redundant model-loading code in gemma2 reward model.

Checklist

• [V] Format your code according to the Contributor Guide.
• [V] Add unit tests as outlined in the Contributor Guide.
• [V] Update documentation as needed, including docstrings or example tutorials.

[merrymercy]

can you also simplify the weight loader of LlamaForSequenceClassification?

[aqweteddy]

Done & verified.

[merrymercy]

Can you fix the lint error?

[merrymercy]

@aqweteddy Thanks for the contribution. It is merged.