fix: ONNX exportability compatibity test and fix

[nimiiit]

fix: Fix for ONNX export of Maxblurpool layer and performance optimization by registering kernel as a buffer so that it doesn't need to be copied to the GPU over and over again.

Description

• What: Converting the pretrained models to ONNX format gives error in the Maxpool layer used in the N2V2 architecture.This is mainly because the convolution kernel is dynamically expanded to a size matching the number of channels in the input in the Maxblurpool layer. But the number of channels should be constant within the model.
• Why: Users can convert the pytorch models to ONNX for inference in thier platforms
• How:
  -- instead of using the symbolic variable x.size(1), explicitly cast it to an integer and make it a constant.
  -- make the kernel as a buffer to avoid the copying to GPU overhead.
  -- add tests for ONNX exportability

Changes Made

• Added:
  -- onnx as a test dependency in pyproject.toml
  -- 'test_lightning_module_onnx_exportability.py'
• Modified: Maxblurpool module in 'layers.py'

Please ensure your PR meets the following requirements:

• Code builds and passes tests locally, including doctests
• New tests have been added (for bug fixes/features)
• Pre-commit passes
• PR to the documentation exists (for bug fixes / features)