Implement dynamic num_classes and resolve dimension mismatch in ResNet18 #91

[alimohmedelsaid26-cell]

I have addressed Issue #91 by making the following improvements to the ResNet models:

1. Added missing FC layer: Implemented the final Fully Connected layer in ResNet18 and ResNet34.
2. Dynamic num_classes: Linked the FC layer to the num_classes parameter to ensure the model adapts to different label sets.
3. Fixed Dimension Mismatch: Set the input features to 512 * 32 to correctly match the output of AdaptiveAvgPool2d((1, 32)).
4. Forward Pass Update: Added torch.flatten to ensure correct tensor shapes before passing data to the linear layer.

Testing

Verified the fix with a test script. The model now correctly outputs the specified tensor shape without the previous dimension mismatch error.

[alimohmedelsaid26-cell]

I've verified the fix locally and it resolves the dimension mismatch error. Ready for review!