Multi modal (Just libraries)

[SignalRT]

This are the changes to build and copy the right dynamic libraries as first step to introduce code changes.

@martindevans, CUDA compilation in Windows doesn´t work. If I see the problem correctly it requires CUDA 12.4. But I'm not sure, and I cannot test it on Windows.

[martindevans]

It looks like llama.cpp no longer offers CUDA 11.7 and CUDA 12.2, instead there are only builds for CUDA 12.4. Should we do the same?

[martindevans]

Test run for CUDA 12.4 only started here: https://github.com/martindevans/LLamaSharp/actions/runs/16153517357, will PR that if it passes, then hopefully this PR should compile too

[martindevans]

I've created a pull request to this branch (see here). It changes over to CUDA 12.4 build.

[martindevans]

I'm going to merge this and kick off a test build.

[SignalRT]

@martindevans, I will be working on the changes during the next days.

[martindevans]

The test run (https://github.com/SciSharp/LLamaSharp/actions/runs/16231321874) mostly passed, just a minor issue with file naming at the end. So we've got a working CUDA build for all your bit whenever they're ready :)