Add U-Mamba architecture to MONAI

[billyluu5704]

🚀 Summary
This pull request adds the U-Mamba architecture to MONAI as proposed in Issue #7472. U-Mamba integrates the Mamba sequence modeling approach with the U-Net framework to enhance medical image segmentation performance.

📦 Changes Introduced
u_mamba.py: Implements the U-Mamba architecture under monai/networks/nets/

init.py: Registers the new model class for MONAI usage

test_networks_u_mamba.py: Adds a basic unit test for model instantiation and forward pass

🔧 Motivation
Adding U-Mamba expands MONAI's support for state-of-the-art architectures and enables research on efficient sequence modeling in medical imaging applications.

✅ Related Issue
Closes #7472

[ericspod]

Hi @billyluu5704 thanks for the contribution but it does need quite a bit more work. The CICD fails do need to be addressed, one being that torch.silu is not an exstant function in any PyTorch version we use. The SSMBlock class is a placeholder but this should be substituted for the actual SSM class from the mamba library, which should be imported as an optional module in your source file. Please do look into these issues and we can return to this PR.

[e-mny]

Hi @ericspod, wanna ask if this is still something we want to add to the repo?

If so, I wish to work on it please!

[ericspod]

Hi @ericspod, wanna ask if this is still something we want to add to the repo?

If so, I wish to work on it please!

Yes we do definitely still want Mamba based implementations in MONAI. If you'd like to contribute something using Mamba we would consider it. @billyluu5704 if you're still interested in finishing this PR I'm happy to review again.