Feature/residual attenton unet

[perctrix]

This pull request extends the UNet implementation with new modular components and factory functions, enabling more flexibility and advanced architectures. The most significant changes include the introduction of attention gates, residual convolutional blocks, and attention-based upsampling, along with new factory functions to easily create UNet variants using these components.

New UNet Components

• Added AttentionGate, UNetResidualConv, and UNetAttentionUpsample modules in components.py to provide attention mechanisms, residual connections, and attention-based upsampling for UNet architectures.
• Updated the __init__.py to export these new components for external use.

UNet Factory Functions

• Added make_unet2d_resconv and make_unet2d_attnupsample functions to quickly instantiate UNet models with residual convolution blocks or attention-based upsampling, respectively.

UNet Class Flexibility

• Modified the UNet class to accept a conv_block parameter, allowing users to specify the type of convolutional block (e.g., standard double conv or residual conv) used in the initial layer.

Testing and Example Usage

• Expanded the example usage in unet.py to demonstrate instantiation and sanity checks for the new UNet variants, including residual and attention-upsampled models.

fixes #4

[ATATC]

No blank lines.