Add pluggable PCOptimizer supporting Adam, AdamW, SGD, and SGD-momentum for local PC weight and bias updates

[Betty987]

Summary

This PR introduces a pluggable optimizer system for PC weight and bias updates.Previously, all layers used a hardcoded SGD-equivalent rule:
ΔW = clamp(lr × gradient, -0.01, 0.01)
Now any layer can use Adam, AdamW, plain SGD, or SGD-momentum, configured via a single optimizer_name key in the hyperparameter config.

Main Changes

• Implementation of SGD,SGD with momentum,Adam and AdamW optimizers and PCOptimizer Class- af7f895
  
• Instantiate PCOptimizer in PCLayer - 64b3bed
  
• Integrate PCOptimizer into step_embed,step_linear and step_attn- 08abe8
  
• Add optimizer config at all entry points in GPTConfig - 3c3a9aa
  
• Pass optimizer config to the PCLayer Instance in:
  • Embedding Layer- 2ff52e
    
  • Attention Layer- 3c472ca
    
  • MLP Layer- 59dc4f5
    
  • Output Layer - 9673be4