Fixes issue with model reconstruction of the upper half of the image & saves model checkpoint in s3

[srmsoumya]

This PR resolves an issue with the model reconstructing just the bottom 50% of the image during validation and stores model checkpoints in the s3 store.

• Adds a shuffle argument to ClayModule that is set to False by default
• Logs model checkpoints to aws s3 bucket

Fixes #156 #138

[yellowcap]

Looks good except for some strange test errors 🐈

[srmsoumya]

@weiji14 I am getting an error with create a conda-lock.yml file with new dependency.

Encountered problems while solving:
  - package pytorch-2.1.0-cuda120py38h1932296_301 requires cuda-version >=12.0,<13, but none of the providers can be installed

For now, I have merged this branch with main, as we need to develop v1 on top of v0.2. We can fix the issues with conda-lock & do a v0.2 release next week.

[weiji14]

@weiji14 I am getting an error with create a conda-lock.yml file with new dependency.

Encountered problems while solving:
  - package pytorch-2.1.0-cuda120py38h1932296_301 requires cuda-version >=12.0,<13, but none of the providers can be installed

Hmm, did you run conda-lock lock --mamba --file environment.yml --with-cuda=12.0? I get the same error you got without the --with-cuda=12.0 flag. For reference, my conda-lock/mamba versions are:

$ conda-lock --version
conda-lock, version 2.5.6
$ mamba --version
mamba 1.5.8
conda 24.3.0

I'll patch this up at #225, and also update the docs slightly under the Note section in https://clay-foundation.github.io/model/installation.html#advanced about re-locking the conda-lock.yml file.