This PR introduces initial support for handling arbitrary matrix shapes by adjusting both the Python model and the Softmax hardware module.

Changes

Python Model:

• Modified to golden model for arbitrary shapes by padding the inputs with zeros as needed.
• Updated to selectively ignore padded input values during the softmax operation

Current Limitations

• Non-zero biases are not yet supported (setting bias to zero for padded values is missing)
• The HWPE version does not yet work correctly. There is most likely a bug
• Only works with ReLU activation (controller for activation is not yet adjusted)
• Feedforward and MatMul only work with one tile (controller is not yet adjusted)

ToDo

• Set bias to zero for padded values
• Fix HWPE tests with padded values
• Adjust controller to handle feedforward and MatMul stages
• Adjust GeLU activation to handle padded values