Reviewed all modified modules and continued, made adjustments and continued with rehaul

[shaltielshmid]

Hi @NiklasGustafsson !
I went over all the modified files in the PR in the dotnet repo. Mostly they were great - I made a slight adjustment here and there (like the wrong class name in nameof or a python naming change).

All the modules that you moved to ParamLess but were still just a wrapper for the unmanaged object - I removed the unmanaged dependency. This took quite a lot of effort for the all the different pooling functions.

In general, for pooling, the functions that return with_indices, I changed all the tuples to be named (Tensor Values, Tensor Indices) to match other usages of that kind of tuple.

I had two debates that I didn't put in yet for Linear and Bilinear:
1] The constructor has an argument hasBias whereas in PyTorch it is just bias. Is that worth fixing?
2] The properties in_features etc. are public properties which can be modified - the question is whether we want to set them to be private set? I wasn't sure since the weight field can be modified.