[RFC] Use each model's official initialization instead of a unified initialization

[sustcsonglin]

Proposal

Use each model's official initialization instead of a unified initialization

Rationale

Related issue https://github.com/fla-org/flash-linear-attention/issues/220 , https://github.com/fla-org/flash-linear-attention/issues/266

[Triang-jyed-driung]

Consider restructuring the code: Use base class FLAModel, FLAForCausalLM which provide the default initialization and forward. Downstream models inherit these base classes and override these functions when necessary.

[zhiyuan1i]

Changed RWKV7 to official initialization