[WIP] LibTorch upgrade

[alinpahontu2912]

Upgrade Libtorch to 2.7.1 and Cuda 12.8

[phizch]

Nice timing! I forked this repo 5 days ago to do the same upgrade. I managed to upgrade it, and all tests passed. But I couldn't figure out how to pack the libtorch packages. I've been trying to figure out how the last few days. But luckily it seems I can soon just update the official TorchSharp packages 😅.

Anyway, I noticed there are a couple of things I did that hasn't been done in this PR. But I don't know if I needed to do them either. But here they are, just in case:

1. The Linux libtorch archive (libtorch-cxx11-abi-shared-with-deps-2.7.1%2Bcu128.zip) had a couple of .so files that aren't included in libtorch-cuda-12.8.proj. I left a comment there.
2. I had to modify THSLinearAlgebra.cpp to get it to compile, even with the torch::linalg:: to torch::linalg_ changes. I had to use c10::string_view instead of the raw char* p value in THSLinalg_cond_str and THSLinalg_norm_str. Since yours compiled without this change it was probably dev-env differences.

[alinpahontu2912]

Hey @phizch, sorry for the delayed reply, I did not receive notification from this PR for some reason and have just seen your comments. I am currently facing some issue with the internal ci when packaging the new nugget files but am actively working on it! Thanks for the support though! If you want to attempt packing them on your device, you can inspect the azure-pipelines.yaml file and check the following steps: Windows_Native_Build_For_Packages, Build_libtorch_cuda_win_Packages and Build_TorchSharp_And_libtorch_cpu_Packages, you might be able to replicate them locally.