Add support to hash via GPUs

[hotcodemacha]

Summary

Closes #201

Introduced a new PyTorch-powered TorchSHA256 hashing engine that chooses a CUDA GPU when available and otherwise falls back to CPU, implementing the complete SHA256 algorithm with explicit 32-bit masking and standard round constants

Exposed a [gpu] optional dependency group to install torch for users who want GPU hashing support

Added tests verifying that the GPU-backed engine produces the same digest and algorithm name as the existing CPU SHA256 implementation

Checklist

• All commits are signed-off, using DCO
• All new code has docstrings and type annotations
• All new code is covered by tests. Aim for at least 90% coverage. CI is configured to highlight lines not covered by tests.
• Public facing changes are paired with documentation changes
• Release note has been added to CHANGELOG.md if needed

[hotcodemacha]

@mihaimaruseac - Please help with review. Thanks

[mihaimaruseac]

I think we actually want the hashing itself to be via a C/C++ module that runs directly on GPU, not via a deep learning library. There is already some work being done by Purdue to hash models on GPU.

[hotcodemacha]

I think we actually want the hashing itself to be via a C/C++ module that runs directly on GPU, not via a deep learning library. There is already some work being done by Purdue to hash models on GPU.

ack

[sandlbn]

I think we actually want the hashing itself to be via a C/C++ module that runs directly on GPU, not via a deep learning library. There is already some work being done by Purdue to hash models on GPU.

ack

Agreed — using a C/C++ GPU module for hashing makes sense, especially if we want to stay vendor-agnostic and flexible with kernel integration. We’re already in touch with Purdue, but we’ll need a broader technical discussion and clear plan to move forward.

[mihaimaruseac]

I think we should close this to not confuse others. Thank you for this PR and the ones that got merged!