Update deps / python to latest supported version - use UV for dep management

[svij-sc]

Python3.9 is reaching has reached end of life on October 31.
This poses some concerns incase there are any security issues w/ python or any of the deps we use as they will likely not publish any patches, etc.
So a need to update python version and deps exist - which is quite heavy handed change as we make use of libraries that are not all available on pypi.

1. Thus, the dependency management is complicated and involves us indexing wheels directly:
   

   GiGL/python/pyproject.toml

   Lines 72 to 83 in a48144b

   	"torch @ https://download.pytorch.org/whl/cu121/torch-2.5.1%2Bcu121-cp39-cp39-linux_x86_64.whl#sha256=3c96b2ec4723e7d97259964ee73e2d6a2bace42511a49005b083ea7be1a0b0ac ; platform_system=='Linux' and python_version=='3.9'",
   	# ===================================== Deps for PyTorch Geometric =====================================
   	"pyg-lib @ https://data.pyg.org/whl/torch-2.5.0+cu121/pyg_lib-0.4.0+pt25cu121-cp39-cp39-linux_x86_64.whl ; platform_system=='Linux' and python_version=='3.9'",
   	"torch-cluster @ https://data.pyg.org/whl/torch-2.5.0+cu121/torch_cluster-1.6.3+pt25cu121-cp39-cp39-linux_x86_64.whl ; platform_system=='Linux' and python_version=='3.9'",
   	"torch-scatter @ https://data.pyg.org/whl/torch-2.5.0+cu121/torch_scatter-2.1.2+pt25cu121-cp39-cp39-linux_x86_64.whl ; platform_system=='Linux' and python_version=='3.9'",
   	"torch-sparse @ https://data.pyg.org/whl/torch-2.5.0+cu121/torch_sparse-0.6.18+pt25cu121-cp39-cp39-linux_x86_64.whl ; platform_system=='Linux' and python_version=='3.9'",
   	"torch-spline-conv @ https://data.pyg.org/whl/torch-2.5.0+cu121/torch_spline_conv-1.2.2+pt25cu121-cp39-cp39-linux_x86_64.whl ; platform_system=='Linux' and python_version=='3.9'",
   	# ===================================== Deps for TorchRec =====================================
   	"torchrec @ https://download.pytorch.org/whl/cu121/torchrec-1.0.0%2Bcu121-py3-none-any.whl ; platform_system=='Linux' and python_version=='3.9'",
   	"fbgemm-gpu @ https://download.pytorch.org/whl/cu121/fbgemm_gpu-1.0.0%2Bcu121-cp39-cp39-manylinux2014_x86_64.whl ; platform_system=='Linux' and python_version=='3.9'",

2. Secondly, to be most secure, this involves us maintaining seperate frozen requirements for each host we support:
   https://github.com/Snapchat/GiGL/tree/main/requirements

3. Finally, there are some other complications of trying to package and host wheels in corp/private registries: https://go/gigl-issue/1434

Proposal to make this easier
Taking these issues into account, we can make our dependencies easier to manage if we adopt uv as part of upgrading our deps / python version: https://docs.astral.sh/uv/

Specifically, the way it helps us is:

1. uv allows us to more easily maintain custom indexes for packages: https://docs.astral.sh/uv/concepts/indexes/
2. uv maintains a lockfile for us that contains information for all hosts / all dependency group resolutions: https://docs.astral.sh/uv/guides/projects/#uvlock
3. The customizability of custom auth paths/certs allows practitioners to more easy package host wheels in corp/private registries: https://docs.astral.sh/uv/concepts/authentication/

Beyond our issues above, uv further improves the developer experience by:

1. Installing and managing our python environment for us
2. It replaces our use of multiple other tools: pip, pip-tools, twine, conda, etc.
3. Its support for workspace allows us to manage GiGL as part of other larger codebases.
4. Container like sandboxing

[svij-sc]

I have done some work to update us to UV.

Helpful guidance:

1. uv helps us manage our python environment, so Instead of doing python -m ...; we will need to do uv run python -m ...; subsequently tools like black can now be executed as uv run black and dont need to be installed at a global level or inside some unmanaged python env. As long as you are in the gigl directory, uv will know how to work.
2. We eliminate the need for using pip to manage/install deps; we also eliminate the need to maintain different requirements files, there is just one now uv.lock

Pull requests:
Note: Each subsequent PR builds off the prior.
Note2: Testing is mostly done in 5, and 6 below (#400 and #401). Further testing will happen post as step 7 and following checks will need to pass before we can merge to main

1. [1/6 Deps Update] Remove Existing Locked Deps. #396
2. [2/6 Deps Update] Udate deps --> migrate uv #397
3. [3/6 Deps Update] Install UV + update dev usage to UV #398
4. [4/6 Deps Update] Update docker images use uv #399
5. [5/6 Deps Update] Update all workflows and actions to use UV #400
6. [6/6 Deps Update] Type check and format #401
7. Scale testing - TBD

[svij-sc]

Things that need to be validated prior to merge to main:

• Testing suite (Unit/Integration/E2E)
• Scale Testing
• Wheel build and publishing
• Doc website building and publishing
• Github Action workflows CI
• Github action workflows comment
• Fossa
• Compiling new protos