CUDA 12 stream destruction fix for multi-gpu systems

[ptheywood]

When using CUDA 12.x in a multi-gpu system, an interaction between flamegpu::cleanup() and cudaDestroyStreams during ~CUDASimulation() was casuing exceptions to be thrown.

• CUDA streams are only valid in the context which was used to create them
• CUDA runtime and driver API methods to query a streams context / status all include UB when passing an invalid stream handle, leading to segfaults
• A CUDA 11.x detection for the context no longer being valid by checking the primary context with the driver API does not successfully prevent the UB being triggered in CUDA 12.x in a multi-gpu system.
• CUcontext can/will return the same handle after a reset, so we cannot "remember" the context that way, but CUDA 12 introduces a method which allows us to compare a unique unsigned long long int which can be used to check the context is a match.
• Also ensures cleanup re-selects the original device at the time of the call, just in case (even though any existing contexts would no longer be valid).

No new tests are required, the existing cleanup tests triggered the issue with multi-gpu + CUDA 12.

Closes #1053

Tested in a multi-gpu linux system with CUDA 11.0, 11.8 and 12.0 with different CUDA_VISIBLE_DEVICES values and on a single GPU system with CUDA 12.0.

[ptheywood]

Changes look reasonable.

Can test on Windows on my office machine if you think necessary.

I think it should be fine, or atleast no worse than the current implementation in that case. I.e. I'm confident in the CUDA 12 behaviour with this patch, I'm less confident about the existing CUDA 11 behaviour.

So I'd lean towards merging this in, and then maybe we should just check it on a windows multi-gpu at some point when convinient anyway, as I doubt we've ran FLAMEGPU2 on a multi-gpu box in windows for quite a while (Unless @mondus' machine has 2 GPUs in currently, rather than just a 3080?).

[Robadob]

Yeah my office machine has the 750ti and titanx on Windows, hence my suggestion of that if we care.